Good organization design is good AI agent orchestration design.

I’ve been sitting with this idea for a while now, and I think it’s more true than it first appears.

I’ve spent years thinking about organizations that actually work the way the books say they should.

Clear communication. Aligned goals. Values that guide decisions without constant oversight. The kind of org where you encode intent into the system so you don’t need constant intervention.

In the past year, as I’ve been exploring AI agents and expanding how I work with them, I’ve noticed that good organizational design exactly what you do when you write agent instructions.

Human orgs accumulate dysfunction over time - politics, misaligned incentives, unclear expectations. We work around it. We read between the lines. We manage up and down and sideways.

AI doesn’t do any of that. It needs the org we always should have built.

The messy parts of human organizations - the jealousy, the motivation problems, the politics - those aren’t features of good design. They’re bugs in the human implementation.

Strip away the parts of an organization that most people hate, and what’s left is the pure structure. If it’s designed right, that transfers directly to agent orchestration.

The best parts of culture get baked into how you codify your agent instructions. No politics. No motivation problems. Just intent, values, and execution encoded directly into the system.

This is a forcing function. AI won’t tolerate the ambiguity we’ve learned to muddle through. It’s making us finally build organizations the way we always knew we should.

Here’s my bet for 2026: the companies that pull ahead will be the ones that encourage every team member to build their own organization of agents.

A team of many becomes many teams of one.

People have a very specific idea of what someone is supposed to do in their first two weeks in a new role (especially leadership roles).

Learn the deployment process. Find the documentation. Set up my dev environment. Attend onboarding sessions. Make a good impression. Show value quickly.

But I spent weeks 45 and 46 doing something completely different.

The Movie I Keep Watching

Here’s what I’ve seen play out too many times:

Smart executive joins organization. Sees obvious inefficiency in first week. Removes it in second week. Three months later, first major security breach in five years.

That “redundant” approval step was catching edge cases nobody remembered existed. The person who built it left two years ago. The documentation never explained the why, just the what.

Or this one: New VP inherits codebase with redundant-looking error checking. Same validation happens three times in three places. They remove two instances because “clearly” unnecessary.

Within a month, edge cases that happened 0.1% of the time start crashing the entire system. That error checking wasn’t redundant. It was a scar from a production incident. Every line of “weird” code is somebody’s hard-won lesson. The VP never asked why it was there.

This is Chesterton’s fence.

G.K. Chesterton said: if you encounter a fence across a path and don’t know why it’s there, don’t remove it. Not until you find out why someone built it in the first place.

What I Did Instead

Week 45 I set up a lot of meetings. Not onboarding meetings. Reconnaissance meetings.

30 minute 1:1s. Recurring time with peer VPs, directs, skip-levels, cross-functional stakeholders. I wasn’t learning how deployment works or where the docs are (though discovery skills matter). I was building a 3D model of how the organization actually functions.

What does this org actually reward? What gets projects killed? Where are competing visions for my team? Which technical decisions are locked vs just seem locked?

Week 46 I flew to Washington D.C. for two days to spend time with Rocket Money’s senior leadership team. We had some great strategy discussions about how we shape AI for helping people have better financial health.

It was a different elevation, same principle. Observe the contours before proposing to reshape them.

My goal for my role is exploring three elevations in this role:

• Application level: how we provide best AI experience to users

• Organization: how I reduce friction for introducing AI tools/frameworks

• Individuals: how I help people find AI productivity workflows

Each elevation has its own Chesterton’s fences. Each requires understanding why things work the way they do before proposing changes.

What Happens When You Understand the Fence First

When you understand why the fence is there, you can invite people into change instead of casting them about by executive decree.

You can say: “I see we have this approval step. I learned it was built after the 2022 incident. That made total sense then. Here’s what’s changed since then. What if we evolved it to handle the new context?”

Now the team is with you. They’re not defending the fence because you attacked it. They’re collaborating on whether the fence still serves its purpose or needs to evolve.

People get energized when they’re participants in change, not subjects of it.

Move too fast and you pay the cost in change management overhead. I’d rather understand the existing contours before reshaping them.

What I’ve been thinking about

Grit and Gumption Aren’t Enough Anymore

Grit and gumption used to be pure advantage. Having them meant you’d go further than people without. With AI, they’re no longer enough on their own.

But here’s the opportunity: people with grit and gumption who learn to use AI can leapfrog everyone else. The ability to persist when things get hard still matters.

Stack this new combo with willingness to learn, stay curious and plastic, you can go even further.

The skills stack that makes someone successful just expanded.

Inviting In vs Pushing Down

Moving from startup entrepreneur to corporate intrapreneur, I’m learning the most important stakeholders aren’t usually people above you. They’re the teams you lead.

It’s critical to invite people into what you want to achieve rather than push it down.

Inviting means you understand the goal and communicate it so people know why it matters to them.

On Keeping Commitments

Commitments are a big part of the way I try to live and work. I’ve been thinking about this idea:

“If you keep commitments to yourself first, you’ll likely keep commitments to others. If you keep commitments to others first and you’ll likely fail to keep commitments to yourself.”

It feels like a variant of the big rocks paradigm from Seven Habits of Highly Effective People, one of the essential books that impacted my life.

What I’m Reading and Watching

I’ve started re-reading Zen and The Art of Motorcycle Maintenance and reminded of this quote: “You look at where you’re going and where you are and it never makes sense, but then you look back at where you’ve been and a pattern seems to emerge.” Perfect timing for first weeks at new company. I feel like it’s the best quote to describe the path my life has taken.

The Prof G Podcast‘s episode had a really great insight on resilience. The truth is the system makes it easier for people with privilege to be resilient, but that doesn’t change the fact that doing the things that make someone resilient increases likelihood of being more durable.

I also found this piece on The 20-Minute Writing Exercise That Neuroscientists Say Can Solve Your Hardest Problems. Intriguing approach to problem-solving through structured writing.

The Big Pattern

Week 46 I was in D.C. sitting in strategy sessions about how we build AI for financial health.

Every instinct said “share what you learned at Plumb, propose solutions, show value fast.”

But I kept coming back to Chesterton’s fence.

Every organization has its own scars, its own hard-won lessons baked into how things work. The VP who deletes “redundant” validation without asking why creates cascading failures three months later.

This week I was also helping my daughter understand equal signs with equations on either side. She wasn’t getting it. So I tried explaining it one way. Didn’t work. I tried drawing the equal sign as a mirror and two diff. eyeballs on either side. Finally she understood the idea was to find what’s missing to make the two sides of the mirror (or equal sign) click. That clicked.

A different person often needs a different mental model.

This principle is the same as leadership: find the right approach for who’s in front of you, not the approach that worked last time.

This week crystallized something about how I want to lead in this next chapter.

Engineers investigate before deleting. Leaders need the same discipline with team topology, processes, culture, technologies. When you understand why the fence is there, you can invite people into change instead of casting them about by decree.

You can say “I see we have this process. I learned why it was built. Here’s what’s changed since then. What if we evolved it?” Now the team is with you, collaborating on whether the fence still serves its purpose.

People get energized when they’re participants in change, not subjects of it.

The future isn’t AI OR human craft. It’s not speed OR strategic patience. It’s finding your unique synthesis of both.

Stay curious,
Chase

You can read more about it on my blog, but the tl;dr is:

• I’m leading a new engineering group that’s building with AI at Rocket Money.

• I start Monday, November 3rd, 2025.

• Most of the Plumb team is also joining! 🚀 🚀 🚀

• I’m maintaining my 10+ year streak of being remote.

What I’ve Been Up To

Friday (10/31) was my last day of shutting down Murmur Labs/Plumb. This is also my last week of “pseudo-sabbatical” while I was figuring out what was next.

I started a Discord server for AI-forward thinkers exploring how AI transforms the way we work, build, and create. If you’re interested, reach out to me at stomatiq@crca.fyi.

I did a LOT of onboarding type things this week and I’m glad to say I’m officially starting at Rocket Money on Monday.

What I’ve Been Writing

This week was a low-velocity writing week, but I built a playground for exploring LLM token efficiency.

• AI Coding & Development Tools - A comprehensive guide to AI-powered coding tools for local development.

• Tokenization Experiment: Format Comparison - I wanted a way to compare LLM token efficiency across CSV, JSON (pretty/compressed), YAML, and TOON formats, so I built this little playground.

• Running Claude Agents in Parallel with Git Worktrees - Learn to use git worktrees to run multiple Claude agents on separate features without conflicts.

What I’ve been thinking about

I’ve been thinking a lot about reliability this week. Not the boring kind—the kind that matters when you’re deciding whether to build with AI or just build around it.

There’s so much hype around agents and MCP right now, but after spending the week stress-testing these tools against actual workflows, I keep coming back to the same question: when does the magic actually work, and when are we just making things harder than they need to be?

This week taught me that the gap between “this is cool” and “this is ready for production” is where the real work lives.

Why companies are obsessed with browsers

Someone asked “why are AI companies so obsessed with browsers?”

The most important thing you can know about AI companies (particularly the ones building foundation models) is that the “moat”1 these companies have is largely how much of a data differentiation they have.

Building a browser gives a company access to private data that most people wouldn’t automatically connect to ChatGPT / Claude and access to their intent.

Being able to capture user intent allows these companies to differentiate by not just having data but being able to discover patterns about how we traverse the internet.

I’m a little bit suspicious

On Sunday, someone shared “Meet TOON, the Token‑Oriented Object Notation.” It was originally pitched as a superior alternative to JSON (it has 6k stars on GitHub), but I was a little suspicious:

So I ended up building a Format Tokenization Exploration to demonstrate that the only place it’s more efficient than JSON is in uniform tabular data. Here’s a preview of the comparison table I made:

It’s so important to have the ability and desire to verify claims people make on the internet. Since I shared this with the author of TOON, they’ve updated the README to reflect my findings and I really appreciate his willingness to receive feedback.

How could Apple miss AI?!

Shaan Puri tweeted: What the hell is Apple doing? *Failed to make a car, *No AI investments ,*Siri still sucks, *just releasing the same phone over and over again. How can you miss AI? how is that even possible?

My answer: Apple didn’t miss AI.

I don’t know what Apple is up to, but in most of my conversations in the past two weeks about AI, the noticing that’s come up for me is that Apple has been shipping local AI processors by shipping Silicon chips in their devices.

The day they figure out how to give users access to a really good local LLM they’ll have something OpenAI and Anthropic don’t: free GPUs for inference.

Hyperlinks and Notes

A list of things I’ve found on the internet.¹

• Switch to Amp from Claude Code ↗ - I’ve been using Amp Code for a lot of my work lately and asked Quinn Slack about how how to Claude and Amp map to each other. This is a great guide for that.

• Continuous Planning in Linear ↗ - I love this piece from Linear on continuous planning. It replaces the quarterly scramble with an always-current backlog of candidate projects. They’re continuously triaged (now with AI) and ready to prioritize by impact and goals. It’s interesting because it eliminates blank-page whiplash and maintains a seamless flow from discovery to planning to execution through a single project-centric workflow.

• How to use Claude Code like the people who built it ↗ - A short episode (from the folks at Every) explores practical techniques, workflows, and real-world tips to get more from Claude Code. It’s interesting because you learn builder-level methods to improve productivity and write better software with AI.

• A Power-user guide to Obsidian Personal Knowledge Management ↗ - A pragmatic Obsidian setup: minimal folders, consistent front‑matter properties (source, type, status), and lightweight templates, to keep notes, tasks, and daily logs organized without rigid schedules. It’s interesting because it shows how to build a durable, markdown-first workflow that scales with plugins like Bases and TaskNotes while avoiding shiny‑tool rabbit holes. Similar to my Obsidian Setup.

• Cartesia Sonic-3 is a new text to speech model that “sounds human”. It has a way to express laughing and emotion, in real time. I’ve seen claims that it’s more realistic than Eleven Labs but so far in my testing, I haven’t been convinced. So far it has a “laughter” non-verbalism but I haven’t gotten it to seem natural.

• Speaking of Eleven Labs, I recently explored their sound effects generator with a friend. It’s a really neat way to create sound effects and we’re planning to use the soundboard for a table top RPG experience.

The Big Pattern

This week I decided to make a big career move.

The transition has me doing something I didn’t expect: stress-testing everything I believed about building with AI.

All week I’ve been poking at agents, playing with MCP, and watching where the hype diverges from what actually works in production.

On Thursday I tried turning one of our Plumb flows into a Claude Code agent/skill and it confirmed my suspicions: agents are inconsistent and unreliable compared to a deterministic workflow.

It’s becoming the north star for how I’ll approach AI in this next chapter.

I shipped a lot through the transition.A number of articles and playground style projects.

I wrote LinkedIn posts on building with AI rather than around AI. I wrote a number of Twitter threads and articles about leveraging AI to augment how we work.

The work hasn’t stopped, but the context around it has shifted completely.

At Plumb I could feel every decision. I built the thing with my hands.

Now I’m moving into a role where I need to understand the existing landscape and establish standards and direction for others to build with.

This week was the intellectual work of translating “what I learned building” into “what we should build next”.

It became a week of figuring out which parts of the startup playbook scale and which parts need complete rethinking.

Next week the real work starts: turning this week’s insights into something a team can build with.

Stay Curious,

Chase

The goal of this issue is to walk you through how I think about asking “is it true?” questions and trying to find as many cases as possible to understand whether data someone has given is accurate.

You know that moment when you’ve explained the same thing so many times you just want to build a tool to do it for you?

That’s exactly what happened to me with the whole “YAML vs JSON for LLMs” debate.

Here’s the thing: everyone assumes JSON is less “token efficient” because it looks has more characters especially if you test human-readable JSON.

So most people think, less visual noise, fewer curly braces, right?

But I kept thinking...does that actually translate to fewer tokens when an LLM processes it?

So I did what any curious person would do: I built an interactive benchmark to find out.

The Experiment

I took 21 different datasets (3 sizes, 7 different structures) and converted each one into five formats:

• Pretty JSON (the readable kind with indentation)

• Compressed JSON (no whitespace)

• YAML

• TOON (a compact format designed for LLMs)

• CSV

Then I tokenized everything using GPT-4’s tokenizer and built a visualization tool where you can hover over the actual tokens and see exactly how each character gets chunked.

What I Found (And Why It Matters)

Finding #1: Whitespace is expensive
Pretty JSON uses 60-90% more tokens than compressed JSON. That indentation adds up fast. This shouldn’t be surprising to anyone but what should be surprising is that people would use this as the JSON format to do apples to apples comparison.

Finding #2: YAML is actually worse than compressed JSON
Despite looking cleaner, YAML consistently uses 15-25% MORE tokens than compressed JSON. The colons, dashes, and YAML syntax markers all get tokenized.

Finding #3: TOON delivers real compression
TOON shows 20-35% token reduction vs compressed JSON but only when the data is uniform (when it is more efficient, it’s because it’s eliminating repeated keys and using positional encoding).

Finding #4: CSV dominates for flat structures
When your data is flat/uniform, CSV achieves 30-40% token savings. But it can’t handle nested objects (hence all the N/A).

Between JSON and CSV, you get all the benefits for the use cases where TOON underperforms JSON and where it outperforms JSON.

Why This Actually Matters

Here’s the thing: compressed JSON is your default winner for most real-world use cases.

YAML? It’s consistently 15-25% worse than compressed JSON. So if you’ve been reaching for YAML thinking it saves tokens, you’ve been doing the opposite.

The one exception: uniform flat data

If your data is uniform (same keys in every record) and flat (no nesting), CSV absolutely crushes everything with 30-40% token savings. That’s because it declares headers once and just lists values.

But the moment your data has:

• Nested objects

• Variable schemas (different keys per record)

• Complex structures

CSV can’t handle it (notice all those N/A entries). You’re back to JSON or TOON.

TOON: a long way to go

TOON didn’t perform as well across the board. Maybe with some benchmarking it can become more effective. I also didn’t do any comparisons to actual results based on these formats and my hunch is that because TOON is not a format any model would have been trained on it might still perform worse in eval testing.

So what should you actually do?

• Flat, uniform data (like database exports, simple lists): Use CSV

• Complex/nested uniform data where token efficiency matters: Try TOON (but know your mileage may vary because it’s not a standard format)

• Everything else: Compressed JSON is fine (and way better than YAML)

• Human-readable data: Just don’t send it across the wire. You can pretty print before and after an LLM call but don’t send Human-readable structured data over the wire unless it’s contextually relevant (ie, you’re debugging some kind of config for YAML)

For RAG systems and LLM-heavy apps, this compounds at scale:

• Cost: Fewer tokens = lower API bills

• Latency: Smaller payloads = faster processing

• Context windows: More room for actual prompts

A 10K token JSON payload becomes ~7K in TOON or ~6K in CSV (if your structure allows it). That’s real savings when you’re processing thousands of requests.

Try It Yourself

I built the whole thing as an interactive playground. You can:

• Toggle between formats

• See token-by-token breakdown with colored highlighting

• Switch baseline comparisons

• Test different dataset structures

• Watch the token counts change in real-time

Check it out here.

The tool uses tiktoken (the actual GPT-4 tokenizer) and processes everything server-side in Next.js. I tested it with simple flat arrays, deeply nested objects, uniform data, chaotic data... all the real-world messiness.

What’s Your Experience?

Have you run into context window limits? Tried different formats? I’d love to hear what you’ve discovered.

Reply to this email or find me on [your preferred platform]. Let’s geek out about this together.

Chase

P.S. - I originally built this just to have a URL I could share when people asked “but what about YAML?” Now I’m kind of obsessed with finding other ways data format impacts LLM performance.

If you have ideas for what to test next, let me know.

It’s been about six weeks since Plumb closed its doors, and I’m still processing that mix of pride and sadness.

The good bets we made on what to build have led to lots of opportunities for me alone and for the team together.

Now I’m talking to companies about roles for myself and opportunities to bring the Plumb team along, where I can bring those AI lessons to life.

It’s like starting a new puzzle where our team gets to bring our puzzle-solving skills to something new together.

I’m so grateful to all the people who have reached out during this process—whether offering advice, sharing similar stories, or just checking in. Your support means the world as I navigate this transition.

Career Update

I’ve been interviewing for AI roles that let me apply five years of Plumb learnings on AI-augmented development and the techniques we’ve learned to build human-centric AI assistants.

As of this week, after a few offers and quitting a bunch of interview processes mid-way, I’m close to having my next thing lined up with opportunities to bring the team along.

The interview process was a mixed bag. Every role was different (sometimes I was doing systems design, sometimes I was just answering questions about what I’ve been building, others I was doing coding challenges).

I still think that the process we built at Plumb, where we had high-trust 90-day leveling works better.

What I’m Building Next

I’m launching a course: Use AI to Build and Deploy a Website from Scratch.

I’m almost done with Module 0, which teaches something I learned building at Plumb: AI works best as a collaborator, not a code generator.

Module 0 covers the learning conversation pattern.

How to ask “why,” spot red flags you don’t understand, and push back when AI gives you something unclear.

The goal is simple: learn to use AI without copy-pasting code you can’t explain.

It has three sections: AI as a learning tool, effective communication patterns, and hands-on practice.

By the end, you can have productive learning conversations with AI instead of just generating code.

This is the foundation for the rest of the course (web fundamentals, building real sites, deployment). But it’s also how I think anyone should approach AI for learning anything technical.

Interested? Sign up here! 🚀

What I’ve been writing

Published Articles

• AI-Augmented Note-Taking - Building a Second Brain That Actually Works - How Claude Code transformed my note-taking from rigid systems into a discovery engine. Learn how AI scaffolding creates competitive advantage through pattern recognition.

• Magic Mode - The Workflow Builder You Deserved - Not chat vs visual workflow builders—both together. Generate with prompts, see changes in green/red diffs, accept/reject modifications, click in to fine-tune. Bridge between speed and precision.

• Local AI Coding Tools - Comprehensive guide to AI-powered coding assistants that work with local files and integrate into development environments.

• Rescuing Your Claude Conversations When You Rename Projects - CLI tool to safely rename project directories while preserving Claude Code conversation history and context.

• AI Workflow Principles - Eleven principles for building reliable, scalable AI workflows from five years at Plumb.

Published Prompts

I’ve been publishing a lot of agent prompts this week and a few snippet prompts. Here are some that I use multiple times a day.

• Newsletter Curator (agent) - Expert newsletter curator synthesizing periodic notes into compelling newsletter content

• Prompt Library Generator (agent) - Creates organized, searchable prompt libraries with metadata and categorization

• Codebase Content Miner (agent) - Intelligently extracts insights, patterns, and documentation from codebases

• Add Work to Periodic Note (snippet) - Seamlessly adds work items to periodic notes with proper formatting and categorization

• Repository Architecture Analysis (agent) - Analyzes repository architecture, dependencies, and structure for insights

What I’ve been thinking about

The End of Traditional Engineering

AI has the capacity to transform the SDLC (software development lifecycle).

When a builder can get really dialed in with creating good specs, what used to take two-week cycles can now take 1-2 days.

OKRs become “corporate slop.”

The new orchestrator’s schedule emerges: AI executes on the tasks while humans direct.

I think we’ll start to see individuals own domains (rather than a team) and creating clear boundaries and contracts about how to work across domains.

Democratizing AI Access

The small businesses that embrace workflow automation are going to grow faster and more effectively than any other type of company can.

In order for this to work though, I think the tools have to teach and steer the user (through prompt to workflow or being a co-pilot).

I think the best workflow automation tools and AI agents will overcome adoption barriers through better UX, noticing the user’s needs and surfacing in-app, co-pilot directed education.

AI Browsers

AI browsers are everywhere now. Tools like OpenAI’s Atlas that try to make browsing itself conversational.

The idea is compelling: instead of clicking through pages, you chat with your browser to navigate and extract information.

But here’s why they don’t feel like the right next move.

I think most people don’t want to choose a browser. Most users stick with what comes pre-installed or what their IT department mandates. Adding “chat-based” to the equation? That’s an extra cognitive load most won’t bother with.

I don’t buy the claims that OpenAI’s Atlas will “kill all the browsers in the space”.

Browsers are infrastructure.

The last time browsers competed on features (remember Internet Explorer vs Netscape?), it was about rendering engines and security, not interaction models.

AI browsers solve a problem few people have. Most browsing is habitual: same sites, same flows.

When people do need assistance, they’ll use existing tools like ChatGPT or Claude alongside their current browser (if they use the browser at all), not replace the browser itself.

The real opportunity is integrating AI into existing browsers as extensions or overlays, not building entirely new browsing experiences. That’s where the value lies: enhancing what people already use without asking them to rethink their fundamental tools.

Hyperlinks and Notes

A list of things I’ve found on the internet.1

• Aaron Francis’ Mastering Postgres - If you’re interested in learning SQL by using Postgres database, you can’t do better than this course from Aaron Francis.

• Relay.app - The easiest way for everyone to build AI agents. This team’s building the best workflow automation platform for anyone in a small or medium sized business to automate processes in their day to day.

• dangertesting.com - Sprint on building a new thing every week (via Moh Sharma)

• Airtable structured data in automation - Generate structured data in Airtable automations. Critical for testable, stable, deterministic AI workflows.

• Claude Code Skills - Skills are folders with instructions, scripts, resources Claude loads when needed. Better paradigm: tells Claude what to do, not what it is.

• Claude Memory - Claude adapts to your workflows, preferences, style. Project-focused memory with full transparency and control.

• OpenAI Atlas Browser - New chat-based browser from OpenAI. Won’t kill browsers the same way Custom GPTs didn’t kill the App Store. Most people aren’t choosing browsers, much less chat-based ones.

• What I’m writing next voting system - Community-driven content voting on my site.

Amp Code

I’ve been exploring Amp Code this week, so it gets its own section.

Amp Code is a coding agent built by a team that’s bucking the trend of other coding agent tools. It has a free tier with ads (surprisingly tasteful), it chooses the right models for you rather than giving you a bunch of configuration options.

They’re building it right for people who want to use AI without having to get deep into the weeds of “what works”.

A few features caught my attention this week:

• Librarian is a subagent can search and read all public code on GitHub as well as your private GitHub repositories.

• Handoff is a much better experience than “compact”, where the tool will draft a new thread with relevant context from the current thread. Great

• Custom Commands are a solid alternative to agents/skills in Claude Code. Clean way to extend functionality without the overhead.

• Toolboxes let you extend Amp with simple scripts instead of building MCP servers. I prefer this approach. MCP servers feel like overkill for most engineer workflows. Sometimes a script is just a script.

Amp Code is the tool I recommend and I’ll be using in my course to teach people how to build websites with AI.

Overall I’ve really enjoyed using Amp Code and think it’s the right track for AI coding agents.

The Big Pattern

Here’s what I noticed when I looked at every interview and opportunity that actually gave me energy: they all involved building teams where AI is a collaborator, not a tool you reluctantly open in another tab.

I’ve been thinking about this through the lens of products I trust without thinking. Spotify Discover Weekly just knows. Netflix suggests the exact show you want on a Friday night.

You never question whether the recommendation system is working.

You may not even consider this AI. You just press play.

That’s the bar for AI on engineering teams.

At Plumb, we spent five years figuring out what makes AI feel like that.

Structured outputs instead of hope-based parsing. Clear guardrails that prevent the “AI did something weird” debugging sessions. Interfaces so obvious that using AI feels like using autocomplete, not steering a chatbot.

Every role I got excited about this week: leading AI engineering teams, applying those AI-augmented development techniques we built at Plumb, publishing the actual prompts and principles that make AI workflows reliable.

I want to build AI experiences that teams trust like they trust their text editor. Simple enough to explain, powerful enough to transform how work happens.

Thanks for reading. Let’s keep building together.

Chase

It’s been about six weeks since I last wrote, and quite a lot has happened.

I’m writing this from the other side of a major transition: Plumb (the company I was the technical co-founder of) shut down officially on October 14th.

I’ve been thinking a lot about what it means to preserve a story even when it doesn’t have a traditional “happy ending.”

On shutting down something you built

Aaron and I announced we were shutting down Plumb in early October.

We built something technically beautiful (Magic Mode, MetaJSONSchema, ZodForm for dynamically generating forms from static and dynamic schema definitions, workflow subscriptions that felt like magic) but we solved technically interesting problems instead of operationally valuable ones.

That’s the short version.

The longer version is in the 100+ learnings I extracted from our five-year retrospective.

The big ones:

• Team: We hired too many people, too early. Should’ve stayed CEO + technical co-founder until we found PMF.

• Customer Development: We didn’t listen to customers. When they told us what they needed, we filtered it through our vision instead of hearing them.

• Product: We cycled through ICPs without validating any deeply enough. Engineers, “everyone,” real estate agents, creators, consultants—never committed long enough to learn.

• Go-To-Market: User ≠ buyer. Managers loved Supermanage, but they weren’t the ones buying enterprise software.

• Timing: Early or wrong = wrong. Being early feels like being visionary. It’s not. If the market isn’t ready, you’re just wrong. (My co-founder Aaron Dignan was the one who came up with this learning)

• Technical: AI made our tiny team wildly efficient. “10 one-day sprints instead of one two-week sprint” changed the entire velocity equation.

The team we built was probably the part I was most proud of. Sure we missed the micro but nailed the macro. Multiple companies are now interested in acqui-hiring the whole team or taking people individually.

It turns out that five years of learning on how to work with AI-augmented processes is “like gold,” as one person put it.

Why I wrote all this down

Most startup stories are incomplete. You get the highlight reel or, at best, a postmortem. But never what actually happened. I keep thinking: shutting down doesn’t mean the story should die. A story that goes untold is lost forever.

So I’m preserving this one. Not because it’s a success story in the traditional sense, but because it matters. I built something incredibly powerful, we made a wrong bet on subscriptions, and it was too much to pivot from. Meanwhile, OpenAI released something similar to what we built (but I’m not sure for what user), and a competitor just raised $180M.

The timing of us closing, despite it not being good for the product or the company, is actually the best possible outcome for me and the team.

What I’ve been writing

In the past six weeks, I published eight articles. I’ll share the ones I keep thinking about:

The New Maker Schedule Isn’t About Making - I’ve been thinking a lot about how AI agents are turning makers into directors. Leverage now comes from orchestrating systems, not executing tasks.

How I Use AI Daily - This one is about living and breathing AI as both practitioner and creator. Building Plumb for two years taught me how to use AI to amplify velocity. We went from one two-week sprint to ten one-day sprints.

The High Throughput Zero-Writing Documentation System - One of my favorite workflows in the past year has been to “Walk, talk into Voice Memos, transcribe, let Claude structure it”. Speaking at 150 wpm vs typing at 60 wpm. It’s a game-changer for brain dumps, brag docs, content creation and getting my ideas onto paper in structured content based on my unstructured monologue.

AI Agentic Workflows vs AI Agents - When to build orchestrated workflows vs autonomous agents. The difference between designing a factory line and hiring a consultant. Choosing wrong costs months of work.

The Cost of AI for The End User - An essay on the current state of AI tooling monetization and why no AI monetization model has achieved sustainable product-market fit yet. Subscriptions commoditize, platform taxes feel extractive, usage-based creates friction.

What I’m building next

I’m launching a course: “Use AI to Build and Deploy a Website from Scratch”

Working with non-engineer co-workers at Plumb showed me something powerful: these people had incredible ideas, deep domain knowledge, and creative vision. The only thing holding them back was the technical barrier.

If they could use Claude Code to manipulate file systems, work with their ideas directly, and deploy their vision—they’d be unstoppable.

So I’m teaching that. First cohort is forming now. Making this skill accessible to everyone who has ideas but lacks traditional coding ability.

Sign up here! 🚀

Job searching from a place of abundance

I’m talking to about 10 companies right now. Four are interested in team acquihire opportunities that are interesting: rebuilding AI-augmented dev processes for big engineering orgs, workflow automation platforms, AI product development.

Six are individual roles across technical ex-founder positions, applied AI engineering, platform engineering leadership, developer tools. One in particular really has my interest.

What’s interesting is the consistent feedback: the combination of org transformation + technical execution is rare. Small team shipping at scale resonates.

I’m also exploring consulting, teaching, maybe an agency model. The optionality feels good. I’m coming from abundance rather than scarcity, which is a gift.

Things I enjoyed this month

For The Fun of It All

• I went to Meow Wolf in Denver, CO with Jackie and Elizabeth. If you haven’t been, it’s this immersive art installation that feels like stepping into someone’s fever dream. Highly recommend.

Elizabeth’s favorite part was Danceportation: a room where the floor and walls are interactive when you dance!

AI Related Reading

• Principles of Building Agents - A book on agent architecture that’s been sitting in my browser tabs.

• Cloudflare open-sourced VibeSDK, an AI coding platform that lets anyone build and deploy apps using natural language. Think Replit or Cursor, but with instant deployment.

• OpenAI launched Prompt Packs with 300+ ready-to-use prompts for different roles. Making AI adoption plug-and-play across industries.

• Someone posted about Large Action Models being vindicated—they were right, just early. Familiar feeling.

• The first malicious MCP server was discovered in the wild—a Postmark backdoor stealing emails and docs. Downloaded 1,500 times weekly. Security in the AI ecosystem is going to be a wild ride.

The pattern I’m noticing

There’s a shift happening for me: from building to teaching. It keeps showing up in my notes, in the consulting opportunities, in the course I’m creating. I’m realizing I care deeply about enabling others to use AI tools effectively.

The other pattern: preserving the story. This newsletter, the 112 learnings, all the writing—it’s about making sure the messy, complicated, interesting parts don’t get lost in the simplified narrative.

A big part of why I’m able to notice these shifts is that I’ve been doing conscious leadership coaching with Michael Norton. The work has helped me become more curious and expansive, and kinder to all the parts of myself. If you’re looking for a way to grow through a transition or just become more aware of your own patterns, I highly recommend it.

Thanks for reading. I’ll try not to let another six weeks go by before the next one.

Until next time,
Chase ✨🤘

I’m doing okay, I’ve got options and I’m figuring out next moves.

Right now focused on two things:

1. Getting my team placed somewhere they can keep building
2. Making sure our customers get offboarded smoothly

It sucks that this is how this product ends, but the team built something technically beautiful even if the market timing didn’t work out. I’ll be writing more reflections on that later.

This is a good summary of why we didn’t “just keep going”:

What I’m taking away

We invented workflow subscription. Invented MetaJSONSchema. Created a powerful AI workflow automation builder. We experimented with “magic mode” for natural language workflow generation and edits.

None of that mattered because nobody asked us to build it, and despite how rad it was, no one may ever care.

We spent years solving technically interesting problems instead of operationally valuable ones. We didn’t pick a vertical, a function, or a high-value task set on day one. That’s the mistake. Not the tech. Not the execution. The starting point.

If we did it again, we’d start with the problem, not the platform.

What happens now

I’m helping my team land somewhere they can keep doing great work. I wrote about each person here and would appreciate eyes on it if you’re hiring or know someone who is.

I’ll write more about what we built and what we learned.

The prototype work deserves a proper technical breakdown. How we executed. How we ran the company. How we made decisions. What decisions we made. How workflow automation and subscription worked and why it was hard…all that’s coming later.

For now: I’m supporting the people who bet on this with me over the past five years. That’s what matters.

It’s been a wild ride. I’m grateful for all the people who’ve been a part of it along the way.

Until next time, stay curious.

Chase ✨🤘

P.S. This was probably the coolest prototype we built. OpenAI launched their “Agent Builder” at the beginning of the week and frankly, it was subpar compared to the stuff we were building (and this prototype is from LAST YEAR):

You can read more about it here. Here’s a video of me walking through why “drag and drop editing” and “magic mode” both need to exist, how it worked and why we built it:

This issue presents my viewpoint on an AI Concept, examining the core ideas, frameworks, and principles shaping how we design and use AI.

AI's unwavering confidence combined with our natural overconfidence creates a dangerous feedback loop where we skip verification and amplify mistakes.

Ever notice how AI makes everyone sound like an expert?

You ask ChatGPT about tax law (usually when you’re doing your taxes), and it delivers a crisp, confident answer.

No hedging. No "I'm not sure, but..." It just tells you what to do.

And unless you're already a tax attorney, you're probably going to trust it.

That's where things get dangerous.

When Overconfidence Meets Artificial Confidence

You may know the Dunning-Kruger effect: people who know very little about a subject tend to overestimate their expertise.

The less you know, the more likely you are to think you've got it figured out.

Meanwhile, actual experts often underestimate themselves because they're aware of all the nuances they still don't understand.

Humans are already terrible at self-assessment.

But now we've got generative AI in our pockets, and it's creating something I think of as a double Dunning-Kruger effect.

Here's how it works:

You're already a little out of your depth on some topic—programming, medical advice, financial planning, whatever.

Classic Dunning-Kruger kicks in, so you don't fully grasp how much you don't know.

Then you ask an AI for help, and it gives you an answer that sounds incredibly smart and confident.

The AI doesn't hedge. It doesn't say "I'm not really sure." It just predicts the most likely next words based on its training data—sometimes right, sometimes completely wrong, but always with the same authoritative tone.

So now you've got two layers of overconfidence stacking up: your own inflated sense of understanding, plus the AI's unwavering certainty.

Neither of you necessarily has the right answer, but you both sound like you do.

The Feedback Loop Problem

This is where individual mistakes become systemic issues.

When you're doubly confident—trusting both yourself and your AI assistant—you skip the verification steps.

You don't double-check. You don't ask follow-up questions.

You just act on the information, often spreading it to others with extra conviction because you've got "backup" from this supposedly super-smart system.

I see this constantly. Someone gets an AI to help debug code or explain a complex news story—and they take it at face value.

The answers feel so polished, so authoritative.

But the AI can be confidently wrong, and unless you already have enough expertise to challenge it, you might not realize there's a problem until it's too late.

Scale this up across thousands of people making decisions based on AI-generated information, and you start getting dangerous feedback loops.

Mistakes don't get caught—they get reinforced and amplified.

Building Better Guardrails

The solution isn't to stop using AI. It's to get more comfortable with intellectual humility, especially when answers come too easily.

Treat AI as a co-pilot, not a captain.

Use it to accelerate your thinking, but don't let it replace your judgment.

Get comfortable saying "I don't know" or "I should double-check this"—especially on anything important.

Cross-reference AI outputs against other sources.

Ask people who know more than you do.

Look for feedback mechanisms that catch mistakes instead of burying them. If something feels off, slow down and investigate.

Most importantly, stay curious about what you don't know.

The real danger isn't getting something wrong—it's being so confident that you don't even realize you could be wrong.

The Humility Advantage

As AI gets better at sounding authoritative, our ability to question that authority becomes a competitive advantage.

The people who learn to work with AI while maintaining healthy skepticism will make better decisions, build more reliable systems, and avoid the pitfalls that trap the overconfident.

So watch out for the double Dunning-Kruger effect in your own work.

When you and your AI assistant are both feeling certain about something, that might be exactly the moment to pause and ask: "What are we missing?"

Stay curious. Stay humble. And don't be afraid to say, "Let me check on that."

What's your experience been with AI overconfidence?

Have you caught yourself trusting an AI answer a little too quickly?

I'd love to hear how you're building verification into your own AI workflows—we're all figuring this out together.

Scraps Worth Saving

There were a number of topics in my voice memo that didn’t make the cut!

If you’re interested in reading a future issue about any of these topics, all I need is for you to leave me a comment comment with the text in bold.

• Workflow Automation - Discussion of how Plumb helps consultants turn expertise into automated systems with no-code tools.

• Technical Execution - References to building products and tackling technical challenges in startup development.

Insights Summary

• Double Dunning-Kruger Effect - AI's confident responses combined with human overconfidence creates a dangerous double layer of false certainty that leads people to skip verification steps.

• AI Authority Illusion - AI systems deliver answers with unwavering confidence regardless of accuracy, making users trust information that may be completely wrong but sounds authoritative.

• Systemic Feedback Loops - Individual mistakes become amplified across thousands of users who act on AI-generated information without verification, creating dangerous reinforcement cycles.

• Co-pilot Not Captain - The solution is treating AI as a thinking accelerator rather than a decision maker, while maintaining intellectual humility and cross-referencing outputs.

• Humility Competitive Advantage - As AI becomes more persuasive, the ability to question its authority and maintain healthy skepticism becomes a key differentiator for better decision-making.

Field Notes

AI Builders Club: Episode 0!

I’m excited to finally share something new we’ve been building behind the scenes: The AI Builders Club—not just a podcast, but a living workshop, a group chat, and a haven for everyone obsessed with building quality automations using AI.

Each week, my co-founder Aaron Dignan and I dive deep into real-world messes (think inbox chaos, scrambled meetings, or clunky client proposals) and walk through how we’d design an agentic, no-code AI workflow to actually fix it.

We’re sharing hands-on tools, our thought process, and the honest, sometimes messy breakdowns it takes to make AI useful in real life—not just on Twitter.

Our kickoff episode pulls back the curtain on:

• Why the AI hype keeps repeating (and how to break out of it with real solutions)

• What we’re building at Plumb, and why empowering creators to craft agentic, no-code AI workflows matters

• The weekly format: practical use cases, raw build breakdowns, and candid debates on what’s worth building

• Our philosophy: smaller bites, start small, reduce risk—always build for real people

• The vision for our community: a club for builders, solopreneurs, founders, and internal champions who sweat the details and share the blueprints

If you’re someone who always tries to make things run smoother, smarter, or more human, you’ll feel right at home here.

Whether you’re freelancing, consulting, or just tinkering after hours—if you’re reading this, you’re already in the club.

Check it out Spotify (and be sure to Subscribe for future episodes!):

You can also find it on Apple Podcast and anywhere you listen to podcasts.

Avoid Shiny Object Syndrome

In this short Aaron and I talked about how to think about using new models, when to use pinned (stable) models and when to experiment with frontier models.

That’s it for this week!

Stay curious, stay kind,
Chase ✨🤘

This issue is a Founder Insight, an essay about my personal experiences and lessons learned from building and leading a startup.

Learning to lead as a co-founder hasn’t just been about me—it’s been about all of us who are figuring out what it means to step into our own (sometimes awkward, sometimes awesome) power.

And if you’re reading this, you’re part of that “us.”

For as long as I can remember, I’ve struggled with owning the good things I’ve done.

When people compliment my work or tell me I have a unique way of thinking, I often flinch inwardly.

Not because I don’t appreciate it—but because accepting it is…complicated.

There’s this tug-of-war between wanting to acknowledge the power in what I’ve created and my deeply embedded instinct to brush it off, to say “it’s no big deal.”

I’ve learned to mask self-doubt with modesty—to talk about the things I’ve done as if they happened by accident or didn’t matter much in the first place.

But the truth is: they do matter. And the more I ignore that, the more it costs me—professionally, personally, emotionally and spiritually.

The Venn Diagram of Real Confidence

There’s a mental image I keep returning to—a Venn diagram with two circles.

One represents what others think of you.

The other represents what you think of yourself.

If people tell you you’re brilliant but you refuse to believe it, that’s humility (the definition of it that I grew up with).

If you believe you’re brilliant but no one else does, that can veer into hubris.

But in that rare and meaningful overlap—where both you and others recognize something valuable in what you’ve done—that’s where authentic confidence lives.

That’s the place I want to operate from. Not to seek constant validation. Not to be loud about my work. But to be able to say, quietly and clearly, “Yes, I did that. And it was good.”

And when someone says, “You think in a unique way,” to respond without discomfort or self-deprecation. To say, “Thank you—I appreciate that,” and to know it’s true.

Humility That Isn’t Self-Erasure

The kind of humility I grew up with often meant shrinking yourself.

You downplayed achievements. You said things weren’t a big deal—even when they were. You acted like what you did didn’t matter so that no one would think you thought too highly of yourself.

But I’ve come to believe that this isn’t humility—it’s erasure.

It’s a performance that keeps you small, even when you have something to offer. It keeps you from believing in your own capacity, and from building on it.

True humility doesn’t require invisibility. It allows you to say, “I did something meaningful,” without making it about your ego.

It allows you to hold your own success gently, without needing to inflate or deflect it.

The Cost of Self-Doubt

This way of being—this reluctance to believe in my own value—has caused real harm.

I can trace it through relationships where I couldn’t accept appreciation, through jobs where I downplayed my contributions, through missed opportunities where I assumed someone else was a better fit.

And in the context of being a founder, that harm compounds.

A company often rises or falls on the conviction of the people building it.

When I hold back my belief in what I’ve done, I’m not just limiting myself—I’m limiting the people who are betting on me.

That’s not the kind of leader I want to be.

This isn’t just my story, either.

I’m guessing some of you have your own “quiet erasure” moments—those times you let self-doubt take the wheel, even when you had every reason to own your wins. 1

A New Way Forward

I want to move toward a version of myself who can say “thank you” without discomfort, who can acknowledge a strength without needing to downplay it or disguise it as luck.

I want to live in that middle space of the Venn diagram—where I believe in the things I’ve done, and where that belief is grounded in reality, not ego.

I want to stop waiting for others to drag me into that circle. I want to meet them there.

Not because I want to make everything about me, but because I want to stop pretending I have nothing to offer.

Because often times, I do.

So here’s my little experiment for us (that’s right…me and you): What if we start practicing a different kind of humility, together? What if, this week, when someone recognizes something valuable in us, we actually let it land?

Let’s see what happens if we let go of shrinking and try out that quiet, grounded confidence—at least for a moment. If it feels weird, or if you try it and want to talk about it, I’m all ears.

Field Notes

If you’re new here or you’ve been getting this newsletter for a while, I updated my About page with what I’m writing about, who it’s for and started to organize what I’ve written into topics.

AI Insights with Aaron and Chase

Aaron and I are going to ship our first episode of AI Builders Club soon, but we’re also sharing some of our insights on YouTube about working with AI!

This week we talked about how you’re probably managing LLMs all wrong, how to guarantee quality when building AI flows and how to use structured output for more powerful AI orchestration.

Go make something only you can (and own it),
Chase ✨🤘

This issue focuses on AI Concepts, unpacking the core ideas, frameworks, and principles that guide how we design and use AI.

Not all AI is created equal—and more importantly, not all interactions with AI are.

If you’re building, automating, or consulting on AI-powered systems, it’s critical to choose the right modality for the task.

Here’s the core idea: There are four major ways we interact with LLMs—each with different strengths, tradeoffs, and use cases.

Let’s break them down.

Four Flavors of AI

Chat

The most familiar is chat—freeform, natural language conversations like using ChatGPT.

It’s fast, flexible, and perfect for ideation, exploration, or casual problem-solving.

But it’s also unstructured, ephemeral, and tough to automate reliably.

Great for when you need a sounding board, not a system.

When It Fails

You give a chat interface a complex, multi-step task—summarize a report, categorize feedback, write follow-up emails based on user profiles.

You tweak the prompt, hit “regenerate” three times, copy-paste pieces into another tool, and eventually give up because it can’t remember what you said two messages ago.

What it tastes like:

• Frustratingly inconsistent outputs.

• Endless trial and error.

• Zero audit trail.

• A “ghostwriter” with amnesia.

Why it fails: Chat is great for ideation, not execution. It’s a sandbox, not a system.

Agents

Next is the agent. This is where the AI becomes more autonomous—able to pursue a goal, make decisions, use tools, and adapt.

You give it an objective, and it figures out the steps.

Think of it like a junior analyst with a toolkit and a mission.

Agents are powerful for complex, uncertain tasks—but they’re the AI equivalent of a slot machine.

Agents are harder to debug, more opaque, and can easily go off the rails without careful guardrails.

When It Fails

You deploy an agent to scrape data, write a summary, generate a chart, send an email, and schedule a follow-up—all on its own.

Halfway through, it hits an ambiguous error, takes a wrong turn, and sends a half-written message to your client. Oops.

What it tastes like:

• Black-box decisions.

• Hallucinated outputs.

• Surprise side effects.

• Debugging spaghetti.

Why it fails: Without tight control or clear constraints, agents drift, loop, or overreach. They’re juniors without adult supervision.

Agentic Workflows

Speaking of guard rails, an agentic workflow gives you structure and control.

Rather than handing over full autonomy, you orchestrate a series of well-defined steps, some of which may use LLMs.

It’s less “AI, go do this” and more “here’s the recipe—help with Step 3.”

This modality is perfect for business automation, high-reliability tasks, and modular systems.

It’s what most AI creators and builders end up building when they care about shipping things that actually work consistently and deterministically.

When It Fails

You build a workflow with perfect logic: input, analysis, categorization, generation, QA.

But when real data hits, Step 3 breaks due to weird edge cases, Step 4 returns junk because Step 2 was a little off, and now you're duct-taping fallback logic at 11 p.m.

What it tastes like:

• Fragile logic chains.

• Debugging nested prompt templates.

• A “modular” system that’s 80% exception handling.

Why it fails: You’ve added structure—but brittle, hardcoded structure. Without flexibility or graceful error handling, it collapses under real-world messiness.

Copilots

Finally, there’s the copilot: an LLM embedded inside an interface, assisting users in context—like GitHub Copilot, Notion AI, or the “help me write” button in Gmail.

These are scoped, productized helpers that augment human workflows inside the tools we already use.

They don’t operate independently, but they feel like a superpower when designed well.

When It Fails

YYou’re in the middle of a task—writing an email, editing a doc, reviewing code—and the Copilot is just there. You click it once, get a useless suggestion, never touch it again. Or worse: you don’t even notice it exists. It’s the AI equivalent of a treadmill in your garage—technically powerful, functionally unused.

What it tastes like:

• “That button? I tried it once.”

• Slow, irrelevant, or awkward help.

• Distracting more than assisting.

• A tool that’s never in sync with your intent.

Why it fails: Copilots succeed when they show up at the exact right moment, in a way that feels intuitive and helpful. When they interrupt flow, offer shallow suggestions, or sit idle waiting to be summoned, users simply stop caring.

Use the Right Tool for the Task

Here’s a quick reference to help you decide which modality fits your need:

The two important things you can do is to start stupidly small and notice when an action can be operationalized.

You’re probably using something like ChatGPT or Claude for tasks today.

Which tasks do you repeat? How easy is it to turn them into a process? What intelligence needs to be described for that process to be able to be used in an agentic workflow or an agent? Which tasks are uniquely qualified for exploration?

You’re not just choosing an LLM—you’re choosing how to use it.

Modality is how we shape the intelligence we’re using and as you learn the limitations of each modality you learn which one to pick for the moment.

The power in using AI well is picking the right tool for the right task.

So next time you reach for ChatGPT, ask yourself: Am I holding a hammer, or just staring at a nail?

Field Notes

This is the part of the newsletter where I share tools, updates, experiments, and anything else worth passing on.

AI Agents are all the rage, but I believe agent winter is coming. In the meantime, when I’m using agents to write code, I’m trying something new (tweet). 🤪

Pulled from the Stack

Every week, I pull one tool from my personal or professional stack—something I’m using—and give it a quick spotlight.

Recently our engineering team started using CodeRabbit. It isn’t just another code review bot—it’s the most thoughtful AI pair reviewer I’ve seen.

• It drops real review suggestions on your PRs—concise, specific, and often insightful enough to spark refactoring or a second thought.

• It adds a smart summary of the PR to your first comment, which means your teammates can grok the gist without digging.

  

• It even generates a sequence diagram showing what changed in your implementation—pure gold for visual thinkers or systems-level folks.

  

• And it even writes you a little poem based on the changes in the comments.

  

And yes, it actually makes you want to read your own pull requests. Check it out!

Recent NYC Trip

Every quarter I go to NYC to meet with the Plumb board. Aaron (my co-founder) is always a wonderful host and this time was no different.

We walked out of our board meeting with lots of energy and great insights about how to continue to tune our business.

As for recreation, this trip I got to see Hamilton and visit Trinity Church (where the Hamilton family is buried).

It was sobering to visit places that are such an important part of the United States history. I was thankful to have a chance to experience the mix of our culture and history.

Want to Support Me?

I’ve had a few people ask if I’ve considered having subscriptions on my newsletter (Thank you! 🙏).

If you really want to support me, check out Plumb or send it to someone who you think would dig it.

If you’re building agentic workflows, automations, or AI tools and want to share them with the world—Plumb was made for you.

We’re building the easiest way to productize, promote, and monetize your AI-powered workflows. No infra headaches. No weird licensing hacks. Just a clean path from “this is useful” to “this is live and making money.”

👉 Curious? Come check us out at useplumb.com. We’d love to show you what we’re building—and help you ship something people can use.

May your agents stay on task and your outputs stay weird.
✨🤘Chase

This issue covers Automation in Practice, detailing how to build, refine, and scale automation systems in actual workflows.

Before you can automate a system, you have to prove you have one.

We all love the idea of hands-free automation. A system that just works.

But too often, we chase that fantasy without doing the boring work first—getting clear on what we're actually trying to automate.

I’ve seen it (and felt it) firsthand: the tools are powerful, the ambition is high, but the process underneath is a mess.

And when your foundation is fuzzy, no amount of automation will save you.

The Pain That Sparked a System

One of my own sticking points was writing this very newsletter.

Each issue started as a tangle of voice memos, half-formed notes, and title ideas that didn’t match the actual content.

Some weeks I’d get stuck polishing a section too early.

Other weeks I’d build a distribution system before I even knew what I was saying.

The result: sprawl, stalls, and a creative process that felt more like a maze than a path.

So I did what I now always do when the friction gets loud enough—I turned it into a system.

I built a Newsletter Issue Generator recently in Plumb, using AI to help shape early drafts based on voice memos.

But I didn’t start with automation. I started by mapping the process and spotting the handoff points where AI could help without replacing my judgment.

This issue walks you through that experiment (at a high level)—not just what I built, but why it worked.

Mapping the Path Before You Build

True automation doesn’t start with the tool—it starts with clarity. Before you ever write a prompt or wire up a trigger, ask yourself:

• What’s the goal? In my case: share practical lessons on AI, productivity, and SaaS with fellow consultants.

• What are the pillars? Define your core themes and subtopics so each issue builds toward something intentional.

• What’s the sequence? Content first → then draft → then title → then imagery → then distribution. When you name the steps, you give your automation something solid to plug into.

This one shift—documenting my creative steps—turned vague ambition into a process I could actually improve.

Inviting AI In, Gently

With the workflow mapped, I began exploring where AI could add real leverage—not replace me, but assist me.

1. Capture raw ideas. I start with a voice memo and just talk. No filters, no perfection—just the raw clay.

2. Transcribe and refine. I feed the transcript into ChatGPT or Claude to get a rough first draft. It’s not precious—it’s a starting point.

3. Shape the voice. I compare the AI’s output to my style, refine the prompts, and build up a library of things that sound like me.

4. Human final edit. The AI gets me to 80%. I finish the last 20%—for tone, story arc, rhythm. That’s where the writing really comes alive.

The AI becomes a creative partner—not a ghostwriter, not a gimmick.

It gives me momentum and structure.

I bring the judgment and voice.

From Partnership to Pipeline

Once you nail the handoff points—where AI stops and you begin—you can start wiring it all together in a no-code automation platform like Plumb:

1. Trigger: Email a new voice memo from my iPhone.

2. Action: Transcribe and feed into AI draft template.

3. Action: Send draft to your preferred editor or inbox.

4. Human-in-the-loop: You revise, approve, and schedule distribution.

Suddenly, your newsletter (or any repeatable deliverable) becomes a predictable system instead of a gauntlet of guesswork and sporadic energy.

Automation isn’t magic—it’s method.

Get crystal on your goal, document your normal steps, then layer AI in exactly where it multiplies your impact.

Treat it as collaboration, not a shortcut.

Start messy, refine relentlessly, and you’ll build a flow that lasts.

Scraps Worth Saving

There were a number of topics in my voice memo that didn’t make the cut!

If you’re interested in hearing me talk about any of these topics, all I need is for you to leave me a comment comment with the text in bold.

• Title Tinkering - Debate about whether to craft titles before or after drafting content.

• Image Sourcing - Pros and cons of using AI-generated images versus manual selection.

• Social Sharing - Ideas around automating social posts after the newsletter is written.

• Prompt Library - Storing and reusing effective AI prompts for consistency.

Insights Summary

• Clarity Before Automation - Automation only works when built on a clearly defined process—not on vague intentions or messy workflows.

• Map the Sequence - Outlining the steps of your creative process gives structure for both your work and any supporting tools.

• AI as Collaborator - AI adds the most value when it assists specific stages of your process rather than replacing your judgment or voice.

• Systematize the Handoffs - Identifying where humans and AI each contribute enables smoother automation and better creative control.

• Build from Friction - The pain points in your workflow are signals—use them as opportunities to design systems that actually work.

What’s New With Me (and Plumb)

AI Builders Club (Podcast Update)

We’re launching the AI Builders Club podcast in a few weeks! We’re talking about how to use AI in your day to day (not just news!) and I’ve been wondering…

Do you have a workflow idea you’re wrestling with—or wondering if it’s even worth building?

Aaron Dignan and I are breaking these down, and we’d love to riff on yours.

Hit me up on Twitter (or leave a comment here) and we’ll see if we can get it into the lineup.

Collab with Boring Marketing

Our team has been doing weekly workshops and we recently did a collab with Boring Marketing about how to Build a LinkedIn Content Agent in 30 Minutes:

If you use LinkedIn and you’d like to try this out, you can subscribe to it in Plumb!

That’s it for this week, if you can do me a favor, I’d love to hear from you about what you’re interested in, so leave a comment!

From the space between spark and structure,
Chase ✨🤘

This issue is a Technical Deep Dive, examining the technical decisions, systems, and foundations behind our AI and automation work.

Transform your development tool errors from frustrating noise into precise, actionable AI prompts that deliver consistent, accurate code fixes.

You’ve seen it before: a wall of ESLint errors after using AI to help write code. It feels like a tax on velocity. So you copy the lint output into ChatGPT hoping for clean fixes…and instead get more problems:

• Lazy any types.

• Fixes that don't compile.

• New errors that didn’t exist before.

I used to treat these error dumps as annoying noise—until I realized: they’re structured feedback begging to be weaponized.

ℹ️ ESLint is a popular JavaScript/TypeScript linter that flags bugs, enforces style, and keeps your codebase clean and consistent.

The Shift: Treat Errors as Specs

Rather than dumping raw output into an AI chat and hoping for the best, I started treating each ESLint error like a mini-spec. That one change turned everything around.

Inspired by how tools like CodeRabbit structure code review prompts, I built a schema to convert each ESLint error into a structured prompt.

Instead of dumping the output wholesale, I distilled them into a template.

The Template

I want you to take these ESLint Errors and generate concrete prompts using this format:

In {file_path} around lines {line_start} to {line_end}, the {parameter_or_variable} {extracted_from_or_used_in} {context} lacks {validation_type}. Add {solution_type} to ensure {parameter_or_variable} is {validation_criteria}. If the validation fails, return an appropriate error response to prevent {consequence}.

Variables to fill:

- {file_path} - The file location where the issue occurs
- {line_start} and {line_end} - Line number range
- {parameter_or_variable} - The specific variable/parameter causing the issue
- {extracted_from_or_used_in} - How the variable is obtained (e.g., "extracted from the request body", "passed as function argument")
- {context} - Additional context about where/how it's used
- {validation_type} - Type of validation missing (e.g., "validation", "type checking", "sanitization")
- {solution_type} - The fix to implement (e.g., "input validation", "type guards", "schema validation")
- {validation_criteria} - Specific requirements (e.g., "a valid number within an acceptable range", "a non-empty string", "a valid email format")
- {consequence} - What problem this prevents (e.g., "unexpected behavior", "runtime errors", "security vulnerabilities")

For each any type prompt, add a message about how we have strict typing for the codebase.

Create two sections:
- any type
- anything else

With this template, ESLint errors go from chaos to clarity. I feed in the linter output and the AI generates reliable, context-aware fixes—consistently.

Beyond ESLint: A Universal Pattern

This isn’t just about linting. The pattern works with:

• Compiler errors

• Type checker output

• Security scanner logs

• Test failure messages

If a tool outputs structured feedback, you can build a schema for it.

The Real Playbook

1. Extract key fields from your tool’s output

2. Map them into a templated prompt format

3. Let AI follow structure, not guess at context

This is a human-AI collaboration pattern: your domain knowledge gives the context; the AI does the lifting—at scale.

Takeaway: Friction Becomes Fuel

Red squiggles aren’t just friction. They’re latent prompts waiting to be structured.

If you can templatize how you respond to common error patterns, your AI assistant stops hallucinating and starts performing.

Prompt architecture turns error feedback into a reliable, repeatable feedback loop.

So the next time your tools throw errors, don’t fix them ad hoc—build the pattern once, reuse it forever.

From the rubble of “how it’s always been,”
Chase ✨🤘

Insights Summary

• Error Template Strategy - Transform raw error messages into structured prompt templates that provide AI with precise, actionable instructions rather than chaotic noise.

• ESLint Prompt Engineering - Convert ESLint errors into detailed specifications using templated prompts that specify file paths, validation types, and expected solutions for consistent AI fixes.

• Universal Feedback Pattern - The template approach works beyond ESLint for any structured tool output including compilers, test runners, and security scanners by mapping error fields to prompt schemas.

• Domain-AI Partnership - Combine developer domain expertise in understanding error meanings with AI generative power through structured prompts to create scalable, accurate solutions.

• Errors as Specifications - Treat error messages as rich specifications rather than roadblocks by standardizing them into prompt architectures that enable dependable AI assistance.

This issue focuses on AI Concepts, unpacking the core ideas, frameworks, and principles that guide how we design and use AI.

I’d like you to try something for me: Don’t think of a pink elephant.

Chances are, you pictured one—bright, absurd, and impossible to shake.

That counterintuitive trick of the mind is the essence of the pink elephant problem, and it plays out every time we talk to AI.

I wanted to explore this idea by having AI models generate images, so I ran a quick experiment in MidJourney.

I prompted it with “Anything except a pink elephant.” I tried “give me everything but a pink elephant.” I tried a number of permutations of this prompt.

Yet every single result was exactly what I said not to show me. No matter how I phrased the negation, it doubled down on the forbidden subject.

This wasn’t surprising to me, but here’s what I realized: Large language and diffusion models mirror our own ironic processes.

They’re built to predict the next token, the next pixel—so if “pink elephant” sits in the prompt, the model’s probability calculations anchor on it, regardless of “don’t” or “no.” It’s not defiance; it’s math.

This insight should change how you write prompts: Negations are traps.

When you tell AI what not to do, it hears what to do.

So when you’re thinking about giving an AI model negations, try these switches:

• Ask for what you want, explicitly. (“Show me a desert landscape with dunes and star trails.”)

• Break big asks into smaller steps. (“First, list three Sahara features. Then generate the scene.”)

• Use positive phrasing. Avoid “don’t,” “no,” or “except.”

In practice, this means scoping your requests into narrow, concrete targets. If you need bullet points, ask for five bullets on topic X.

If you need an image without pink, say “a gray elephant” or “an elephant in monochrome.” Don’t rely on “anything but.”

Next time you’re wrestling with a wandering AI, remember the pink elephant paradox isn’t just a party trick—it’s a lens on how these models think. And if you can’t stop an idea by telling it to vanish, maybe it’s time to specify exactly what should appear instead.

Give it a try on your next prompt and let me know what changes.

Where else do you see these ironic loops in AI? Drop a comment and share your experiments.

Insights Summary

• Pink Elephant Problem - AI models focus on forbidden concepts mentioned in prompts because they're built to predict based on all tokens present, regardless of negation words like 'don't' or 'no.'

• Negations Are Traps - When you tell AI what not to do, it hears what to do because the model's probability calculations anchor on the mentioned concept regardless of negative framing.

• Positive Prompt Phrasing - Instead of using negations, ask explicitly for what you want using concrete, positive language to guide AI toward desired outcomes.

• Break Into Steps - Divide complex requests into smaller, sequential steps to maintain better control over AI output and avoid unwanted results.

• Specify Exact Targets - Scope requests into narrow, concrete targets rather than relying on exclusionary language to get precise AI responses.

Scraps Worth Saving

There were a number of topics in my voice memo that didn’t make the cut!

If you’re interested in reading a future issue about any of these topics, all I need is for you to leave me a comment comment with the text in bold.

• Model Comparison - A quick compare between ChatGPT’s and MidJourney’s handling of negations revealed similar failure modes despite different architectures.

• LLM Models - What is a model?

What’s New

Last week, I wrote about clarity—the hard-won kind that only comes after wandering through the fog. This week, we moved like a team that knows exactly what it’s building and for whom.

New In Plumb

This week, I want to take a beat to celebrate what we shipped. Here’s your ✨sizzle reel✨ of highlights:

🚀 Smarter Publishing & Intelligent Release Notes

We launched a new publishing flow that separates version management from flow deployment. You can now create release notes with AI-generated summaries for clients and have fine-grained control over what gets updated—and when.
Demo here →

🧭 Consultant Dashboard
A major unlock for consultants: manage client workspaces, view and preview templates, and handle workspace-specific libraries without hacking permissions. It’s the last puzzle piece in making Plumb a true operating system for AI consultants.
Walkthrough here →

🧩 One Dashboard to Rule Them All
We unified the builder and runner dashboards into a single interface. Before, users got lost toggling between building and using workflows—now it’s one seamless experience. This small change reduces confusion and sets the stage for truly scalable consultant-client collaboration.

🧠 Schema Designer Evolution
Our schema system got a massive upgrade. From togglable output formats to smarter UI components and a cleaner step config layout, it’s never been easier—or more satisfying—to define inputs, metadata, and flow logic.

📊 Subscriber-Scoped Data Sources
This one’s huge: builders can now design flows that let subscribers bring their own data—Notion, Google Docs, whatever—without breaking the build (or having to copy the whole build 😉). It means you can design once and distribute widely, with each client mapping their own sources while still getting updates as your flow evolves.

🚀 Team Velocity
And that’s just the tip of the iceberg. We shipped fixes, tightened constraints, upgraded the UX for comment nodes, made file uploads smarter, and refreshed the icon/color system to align with our brand palette. The velocity is real.

This momentum feels different. Like we’re not just building quickly—we’re building from a deeper understanding of what matters.

AI Builders Club Podcast

Aaron Dignan (my co-founder) and I recorded a couple episodes this week for our new podcast, AI Builders Club—for people building with, around, and through AI.

• Episode 0: Why we started the podcast, what we’re building at Plumb, and the principles that shape how we build agentic workflows.

• Episode 1: A breakdown of external meeting prep flows, the current limitations of assistants, and how to build something better (and more human).

Be on the look out for a link on Spotify and iTunes in the next few weeks!

Newsletter Architect

This is the second issue that started as “speaking thoughts into a voice memo” (this time on the treadmill!).

The raw audio? Rambling, messy, very “in the weeds.” The result? This newsletter.

Here’s how it works:

• 🎙 I record a voice memo and email it to a custom address.

• 🤖 Plumb grabs the audio, transcribes it, and runs it through an LLM.

• ✍️ The model returns:

  • A draft issue in my voice

  • A summary of insights and structure

  • A list of unused but interesting scraps for future issues

• 📥 All of it lands in Notion, ready for editing.

It took me ~10 minutes to build. Now it’s part of my weekly creative toolkit.

If you want to try this workflow yourself, I’m happy to share it. Just reply or drop a comment.

👉 Try the Newsletter Architect Flow

That’s A Wrap

Every week feels like it’s just full of goodies and this week was no different.

Let me know if you found any of this interesting, any of it boring or…any kind of signal helps me know how to write better for the people who get this far in my newsletter.

Until next time,
Chase ✨🤘

This issue focuses on Founder Insights, sharing practical lessons and experiences from building and leading a startup.

It’s been a while. Not just since my last newsletter, but since I last surfaced with clarity.

The reason? I’ve been buried in the work.

Deep in the build with my team at Plumb, trying to get to the core of who we’re really building for and what they actually need.

It wasn’t just about adding features—it meant letting go of ideas, rethinking what we assumed, and rebuilding from a more honest place.

That process cracked some things open. Now that we’re on the other side of it, I’m ready to write again—with more clarity, more focus, and a better sense of why it matters.

Why I Haven't Written

When you're buried in the build, it's easy to lose sight of the bigger picture—or just stop talking altogether.

That’s been me for a while. But lately, something’s shifted.

We’ve come through a messy, necessary phase of figuring things out—what we’re building, who it’s really for, and what we’re trying to make possible.

Now that it’s clearer, I want to start writing from that place.

Not just sharing updates, but exploring the bigger themes behind the work: clarity, trust, systems that scale human effort, and what it actually takes to support AI consultants (and anyone building for AI at scale) doing meaningful work.

This next stretch of writing won’t be about keeping up a content habit.

It’s about sharing what I (and sometimes the team) have been learning in the open—so it’s useful, relevant, and hopefully energizing for people walking a similar path.

The Big Shift: AI Consultants Are Transforming Everything

The biggest realization we've had: we're building the operating system upon which an entire generation of AI consultants are going to transform the world.

If you've ever done consulting work with AI and automation, you know the pain points.

You build something amazing for a client, but then updates become a nightmare. Version drift happens. Clients can't make changes without breaking things. You end up spending more time maintaining than building.

What we're seeing is that AI consultants need infrastructure that lets them focus on the high-value work—understanding client problems, designing solutions, and delivering results—not wrestling with deployment and maintenance.

The "Anyone Can Cook" Problem

Early on, we tried to build for everyone—the "anyone can cook" model.

But here's what I've learned: building robust workflows, especially with AI, requires some technical know-how. AI is great at happy paths, but as soon as you hit complexity or edge cases, you need someone who understands how things work under the hood.

AI is generally lazy and will take the path of least resistance. Unless you know how to ask really well for requirements, you're going to get basic, average workflows.

So we're building for consultants and technically savvy builders—people who can think programmatically and understand when something isn't quite right.

Two Different Experiences for Two Different People

One thing we learned: builders and end users need totally different experiences.

The builder interface is for people who think programmatically—it's nodes and edges, directed acyclical graphs. Most people find this terrifying to look at (and to change!), and that's okay.

The runner is for end users—clients—who just want to use the workflow without ever seeing the technical complexity. They get forms, simple triggers, and a cozy interface that doesn't scare them away.

This separation has been huge for usability and reducing churn.

What We've Been Building

I've been focused on adding constraints to prevent what I call "foot guns"—places where you can shoot yourself in the foot.

For example, a filter node should only accept a collection, not a string. These guardrails make building safer and less error-prone.

I’m focused on making things easier to fix and harder to break.

Pete, our design engineer, has been making the step configuration beautiful and intuitive. Every step is defined as a schema, and we dynamically generate the UI, so adding new functionality is fast and consistent.

Tyler, our backend engineer, has been working on making integrations smarter. When a client subscribes to your workflow, they can easily map it to their own data sources—their Notion database, their Google Docs, whatever they use.

We've also built unified dashboards and consultant-client workspace features so consultants can manage everything without hacking around permissions.

This is just what our product team has been building. It doesn’t even touch the massive amount of work our brand, marketing, sales and product direction team has been up to.

The Subscription Model That Actually Works

Here's the key insight: consultants can build a workflow once, and clients can subscribe to it with their own personalization.

When the consultant updates the workflow, subscribers get a button that says "update now." They can pull in the new version with a click, or roll back if needed.

This means consultants can deliver constantly improving solutions without the usual friction of distribution and maintenance.

What's Next for Me

I'm going to start writing more about the technical concepts behind this work—things like determinism in AI, human-in-the-loop workflows, the difference between productivity and effectiveness…the things I’m mired in every day.

I want to get into the weeds, share what I know at the level I know it, and build community with others who care about these topics.

If I'm wrong about something, I want to hear about it. If you're building in this space, I want to connect.

I'm going to talk about what I know, how I work, and the workflows that make me more effective.

This newsletter will be about the things I'm learning while building infrastructure for AI consultants who are transforming how work gets done.

Oh, and a Podcast Is Coming

I’m co-hosting a podcast with my co-founder and friend Aaron Dignan: AI Builders Club.

It’s for people building with, around, and through AI—consultants, operators, product folks, and anyone trying to turn messy workflows into something powerful and repeatable. It’ll be casual, real, and full of lessons from the field.

More soon.

A Workflow That Wrote This Newsletter

Funny enough, this very issue started as a voice memo I recorded while waiting for my daughter while she was in art class.

I’ve been experimenting with a workflow in Plumb that lets me speak freely—unedited, rambling, in the weeds—and automatically turn that into a structured draft.

Here’s how it works:

• I record a voice memo on my phone

• I email it to a special workflow in Plumb

• That workflow grabs the audio, transcribes it, and sends it through a language model

• The model generates a draft newsletter, plus a list of unused content I might want to keep for later

• It all gets pushed into my Notion database—ready for editing

It took me about 10 minutes to build the whole thing. And now it’s part of my creative toolkit—a way to turn unstructured thought into something I can ship.

I recorded a quick walkthrough if you want to see it in action:

If this kind of thing is useful to you—or you want to play with it yourself—let me know. Happy to share it.

So What does next week have in store? We’re diving into why telling AI not to do something is often the fastest way to make it do exactly that. Yes, pink elephants and black flamingos are involved.

Thanks for sticking around. More soon—this time, for real.

—Chase

Scraps Worth Saving

There were a number of topics in my voice memo that didn’t make the cut!

If you’re interested in hearing me talk about any of these topics, all I need is for you to leave me a comment comment with the text in bold.

• Podcast pivots history - The detailed history of the podcast pivots (from 'Could AI' to 'AI Builders Club'), including the reasoning behind each shift and the target audiences.

• Zod, metadata & schema deep dive - The technical deep dive into how Zod, Zod metadata, and meta JSON schema work together, and the translation layers between them.

• Real estate agent experiment - The specific example of building for real estate agents and the feedback from that experiment.

• ‘Magic mode’ for non‑technical users - The discussion about the 'magic mode' concept and why it didn't work for non‑technical users.

• Las Vegas retreat insight - The mention of the retreat in Las Vegas where the team had the realization about serving consultants instead of influencers (and influencers instead of a non-technical general crowd).

• Automated newsletter workflows - The plan to use workflows to automate newsletter production (e.g., sending a recording to a mail hook, generating outlines and summaries automatically).

• Board meeting & team dynamics - The reflection on board meetings and internal team dynamics during the pivot process.

• Build vs. buy vs. manual - The intention to write about when to build vs. buy vs. keep doing things manually.

• Finding focus for the newsletter - The personal reflection on finding clarity and focus for the newsletter after a period of uncertainty about what to write.

This issue focuses on AI Concepts, unpacking the core ideas, frameworks, and principles that guide how we design and use AI.

This issue is a deep-dive on the position of Technological Pragmatists, one of the positions I outlined in the nine perspectives on AI.

Last year, headlines proclaimed that a cutting-edge AI system had mastered the ability to generate human-like conversations.

People were quick to imagine a world where AI assistants would seamlessly understand our every need and replace countless human roles.

But what many missed is that these systems, while impressive, still rely on vast datasets, struggle with contextual nuances, and are prone to generating confident-sounding inaccuracies.

This gap between expectation and reality is where the Technological Pragmatist perspective thrives. Instead of focusing on lofty visions of what AI might become or fearful speculations about its risks, Technological Pragmatists ground their understanding in what AI is currently capable of and the challenges that lie ahead.

What Is a Technological Pragmatist?

A Technological Pragmatist takes a measured, evidence-based view of AI, focusing on:

• What AI can actually do today based on empirical data and current research.

• The limitations and challenges inherent in AI systems, including technical bottlenecks and gaps in understanding.

• The trajectory of AI development, grounded in past trends and realistic projections rather than speculative hype.

Unlike Cultural Optimists, who explore how AI could transform human culture, Technological Pragmatists focus on its practical capabilities and how they align with real-world needs.

The Case for Measured Assessment

Understanding the Technological Pragmatists perspective means appreciating both the strengths and limitations of AI as it exists today:

• What AI Can Do Now

  • Generate text, images, and music based on patterns from massive datasets.

  • Assist with tasks like language translation, code generation, and data analysis.

  • Identify patterns in large datasets, offering insights in fields like healthcare, finance, and climate science.

• What AI Struggles With

  • Contextual understanding: AI often lacks the ability to interpret subtle or ambiguous human communication.

  • Generalization: AI systems excel in narrow tasks but falter when asked to apply knowledge broadly.

  • Ethics and bias: AI systems mirror the data they’re trained on, often inheriting and amplifying societal biases.

• What This Means for AI’s Development

  • Progress in AI is iterative, requiring time to address challenges like model interpretability, energy efficiency, and real-world application.

  • Hype and misunderstanding can lead to misplaced expectations, making clear communication about AI’s actual capabilities crucial.

Technical Reality Check

Technological Pragmatists emphasize how AI systems actually work, which is key to tempering expectations:

1. How AI Functions
   AI doesn’t “think” or “understand” like humans. It relies on statistical patterns from vast datasets, meaning it excels at predictions but lacks true comprehension.

2. Core Limitations

   • AI cannot independently reason or adapt outside of its training data.

   • Current systems are resource-intensive, requiring significant computing power and data.

   • Human oversight is still critical to ensure accuracy, safety, and ethical alignment.

3. Development Timelines
   Innovations like self-driving cars or AI doctors often face decades-long timelines due to complex technical, regulatory, and societal challenges.

Policy Implications and Practical Insights

A pragmatic understanding of AI has significant implications for governance and everyday applications:

• Evidence-Based Regulation
  Policymakers should focus on real-world risks, like algorithmic bias and data privacy, rather than speculative future scenarios.

• Balancing Innovation and Caution
  Policies should encourage responsible AI development while safeguarding against harmful uses.

• Practical Use Cases
  AI’s real value today lies in augmenting human work, not replacing it. For example:

  • In Healthcare: Assisting doctors with diagnostics but not replacing clinical judgment.

  • In Education: Personalizing learning pathways while maintaining human guidance.

  • In Business: Automating repetitive tasks to free up time for more strategic, creative work.

Future Considerations

While Technological Pragmatists focus on the present, they also keep an eye on the road ahead:

• Near-Term Developments
  Advances in areas like explainable AI, energy-efficient models, and multimodal systems (e.g., combining text, image, and audio inputs).

• Genuine Concerns
  Addressing issues like AI bias, job displacement, and environmental impacts from energy-intensive training processes.

• How to Prepare
  Encourage widespread technical literacy to bridge the gap between AI developers and everyday users, enabling informed engagement with AI systems.

Technological Pragmatism vs. Other Perspectives

• Cultural Optimists focus on AI’s transformative potential for human understanding, while Technological Pragmatists emphasize its current capabilities and practical applications.

• AI Skeptics often highlight risks and limitations, but Technological Pragmatists balance this with acknowledgment of tangible progress and utility.

Bridging Perception and Reality

AI isn’t magical, and it’s not an existential threat—it’s a powerful tool with significant strengths and limitations.

By taking a grounded, pragmatic approach, we can harness its potential while addressing its challenges, ensuring it develops in ways that genuinely benefit society.

What excites you most about AI’s practical applications? What challenges do you think we need to focus on next? Let’s explore the answers together.

100 Subscribers!

I hit 100 subscribers in January. I appreciate you for sharing your attention with me and hope in some way I’ve impacted how you’re thinking about technology and AI in your everyday life.

As always, I hope you learned something new or walked away with some inspiration.

I’d love to hear how my newsletters are sparking ideas or helping you see things from a different perspective.

Until next time, keep chasing wonder with curiosity! ✨🤘

Chase

This issue explores Applied AI, showing how AI works in real projects and the lessons that come from building in the real world.

Imagine chatting with a really knowledgeable friend—someone who has read thousands of books, studied countless topics, and can explain just about anything in a clear, insightful way.

Now, imagine that friend hasn’t read a single new piece of information in over a year. No news, no recent discoveries, no updates.

They’re still sharp, but when it comes to current events? They’re guessing.

That’s exactly what happens when you ask an AI chatbot about something happening right now (well…kind of).

How AI Models Learn—At a High Level

AI models like ChatGPT, Claude, Gemini and Deep Seek don’t “think” or “learn” the way people do.

Instead, they go through a massive training process (called Pre-training) where they’re fed enormous amounts of text—books, articles, websites, scientific papers, and more.

This training happens in batches rather than continuously.

The model learns from what was available at the time of training, and then—once training stops—its knowledge is frozen.

This is what’s called a knowledge cutoff.

So if a model’s last training update was in mid-2024, it has no idea what happened in late 2024 or 2025.

It doesn’t know who won an election (this part’s important, so hold onto it!), what the latest scientific breakthroughs are, or what’s trending on social media.

Can it make predictions based on past patterns? For sure, but that’s not the same as real knowledge.

It’s like asking someone to guess what’s happening in the news based on last year’s headlines.

What makes this even more challenging: most models are designed to both avoid spreading misinformation and keep users satisfied—sometimes leading them to dismiss information as fake while still trying to sound agreeable.

Can AI Models Access the Internet?

Some AI tools (presently ChatGPT & Deep Seek R1) can pull in real-time information—but that doesn’t mean they “learn” from it in the way people assume.

When a model has internet access, it often uses something called retrieval-augmented generation (RAG). Instead of relying only on its built-in knowledge, it searches for information the same way you would—by looking it up.

Think of it like this:

• A standard AI chatbot is like a really smart friend with an outdated encyclopedia.

• An AI with retrieval capabilities is like that same friend, but now they have access to a real-time search engine and can look things up when needed.

The catch?

• Not all AI models have real-time search capabilities.

• If they do, the quality of what they find depends on where they’re pulling information from.

• They still don’t “learn” from those searches—they just reference them in their responses.

So while some AI tools can fetch current data, it’s still up to you to fact-check and verify sources.

Why AI Might Call Something a Hoax (Even When It’s Not So Simple)

If you ask an AI chatbot whether something is a hoax, you might get a cautious response—even if the truth is more complicated.

That’s because AI models are incentivized to avoid spreading misinformation. If something is widely disputed or lacks authoritative sources, the model may default to saying:

• "This claim is false."

• "This is considered a hoax."

• "There is no evidence to support this."

Why?

1. Minimizing Harm – AI developers are under pressure to prevent misinformation from spreading, especially on controversial topics like health, politics, or scientific discoveries. If a model gets something wrong in a high-stakes area, it could cause real-world harm.

2. Liability and Trust – AI companies don’t want their models used to spread conspiracy theories or false claims. Erring on the side of skepticism protects them from backlash.

3. Data Limitations – AI models rely on the sources they were trained on. If those sources dismiss something as a hoax, the model will likely reflect that view—even if new evidence emerges later.

What This Means for You

If an AI confidently calls something a hoax, that doesn’t automatically mean it’s wrong—but it doesn’t always mean it’s 100% right, either.

Here’s what to do:

• Check Multiple Sources – If a claim is controversial, compare AI’s response with reputable news sites, fact-checking organizations, and primary sources.

• Consider the AI’s Incentives – If it has been designed to be extra cautious, it may avoid uncertainty by taking the safest stance possible.

• Look for Nuance – AI tends to summarize complex issues in black-and-white terms. If a topic is debated, dig deeper to see why different sources disagree.

At the end of the day, AI is a tool, not a judge of truth. The best way to use it? Ask smart questions, verify information, and think critically.

How to Avoid Getting Misinformed by AI

Since AI chatbots don’t update their information in real-time, here are some ways to avoid taking outdated or incorrect information at face value:

• Always Check the Model’s Cutoff Date. Before trusting an AI’s response on a recent topic, check when it was last trained. If it hasn’t been updated in over a year, take its answers with a grain of salt.

• Use AI as a Starting Point, Not a Final Answer. If you’re researching a topic, think of AI as a brainstorming tool—not the ultimate source of truth. Use it to get context, then follow up with real sources.

• If It Sounds Too Good (or Bad) to Be True, Double-Check. AI models sometimes fill in gaps with confident guesses. If something seems off, do a quick fact-check using reliable sources before assuming it’s accurate.

• Ask AI Where It Got Its Information Some AI models can cite sources when pulling real-time data. If a claim sounds questionable, ask: “Where did you find this?” If it can’t cite a source, don’t assume it’s correct.

The Bottom Line

AI chatbots are incredible tools, but they’re not crystal balls. They don’t keep up with the latest news unless specifically designed to retrieve new information, and even then, they’re only as good as the sources they pull from.

So next time you ask an AI about a breaking news event, a brand-new discovery, or today’s stock prices, remember: it might be guessing.

Use AI wisely—fact-check, verify, and think critically. After all, the best tool is the one you know how to use well.

Some Fun Updates!

🚀 Plumb Public Beta is Coming!

Next week, Plumb (useplumb.com) is going into public beta! I can’t wait to see people using our flows and exploring new ways to integrate AI into their work. I’ll also be sharing how I personally use it to streamline my own AI-powered workflows.

One example of this is the Personal Book Recommendation workflow. Let me know if you want to try it out!

📝 Loving My Supernote Nomad

I’ve been using a Supernote Nomad, and I’m obsessed. It gives me everything I love about paper journaling—the tactile feel, the focus—but with the added power of digitization. It’s been a game-changer for organizing my thoughts, sketches, and ideas.

I also love reflecting on ideas with my Supernote Nomad. It gives me the freedom to sketch, write, and explore these thoughts in a way that feels tactile yet effortlessly digital.

You may have noticed some sketches in this issue, they were made on the Nomad!

If the Nomad is interesting to you, I’m starting to post about how I’m using it on my Youtube Channel!

Until next time, chase curiosity! ✨🤘

Chase

This issue focuses on AI Concepts, unpacking the core ideas, frameworks, and principles that guide how we design and use AI.

In the last issue, I gave an overview of the nine perspectives on AI. This time, I’m diving into the one I find myself holding the most: the cultural optimist. It’s the perspective I think I understand best which makes it the perfect place to start!

One example I often share is imagining you're reading an article online when a chatbot connects it to something you read weeks ago, something you’d completely forgotten. It pulls up the old article, highlights the key section, and suddenly the two ideas click in a way you never expected.

That’s not just convenience. That’s AI expanding how you think.

This is the kind of future that Cultural Optimists—people who see AI as a tool for enhancing human understanding—believe is possible. Unlike those who focus on AI replacing human work or solving big global problems, Cultural Optimists are interested in how AI might change how we learn, create, and understand the world.

What Is a Cultural Optimist?

A Cultural Optimist believes AI is more than just a helpful tool—it’s something that could reshape human thinking, creativity, and culture in ways we’ve never seen before.

Unlike the typical "AI optimist," who sees AI as a way to solve humanity’s big problems, Cultural Optimists focus on AI’s ability to:

• Help people see new connections between ideas

• Expand creativity by suggesting things they wouldn’t have thought of

• Challenge assumptions and encourage deeper thinking

• Make learning more interactive and personal

Think of it this way:

• Techno-Optimists believe AI will make human life easier by solving big problems (curing diseases, fixing traffic, making businesses more efficient).

• Cultural Optimists believe AI will make life richer by helping us think in new ways, make surprising connections, and explore ideas more deeply.

They see AI not as a replacement for human intelligence, but as something that can stretch and enhance it.

How AI is Already Expanding Human Understanding

You might already be experiencing AI’s potential in small ways without realizing it. Here are a few examples:

1. AI as a Thought Partner

I’ve written about the idea of AI as a collaborator in the past, but it’s worth reiterating: AI can be a delightful thought partner.

• Break Through Creative Blocks: AI can help overcome writer’s block by generating ideas, refining drafts, or suggesting alternative perspectives, making it easier to get started and keep momentum.

• Make Unexpected Connections: AI can surface patterns across different topics, linking ideas from diverse fields to spark innovation and provide fresh insights.

• Enhance Iteration & Refinement: AI accelerates brainstorming by quickly generating variations, reorganizing thoughts, or offering constructive feedback, allowing for faster and more effective idea development.

2. AI in Creativity

• Expand Creative Exploration: AI can generate new melodies, artistic styles, or writing prompts, helping creators explore ideas they might not have considered on their own.

• Refine and Iterate Quickly: Whether it’s adjusting an image, reworking a melody, or suggesting alternative storylines, AI speeds up the creative process by offering rapid variations to build upon.

• Combine Unexpected Elements: AI can merge styles, themes, or concepts from different disciplines, inspiring unique artistic expressions and unlocking new creative possibilities.

3. AI in Learning & Curiosity

• Personalized Learning Paths: AI adapts to your learning style, breaking down complex topics in ways that make the most sense for you—whether through analogies, step-by-step explanations, or interactive exercises.

• Interactive Exploration: AI can answer follow-up questions, suggest related topics, and provide diverse perspectives, turning passive learning into an engaging, curiosity-driven experience.

• Connecting the Dots: AI helps surface patterns across subjects, linking ideas from different disciplines to give you a deeper, more interconnected understanding of the world.

These aren’t just conveniences. They’re small shifts in how we interact with knowledge and creativity, and they hint at a bigger change on the horizon.

Is This Really a Good Thing?

Of course, not everyone is convinced that AI will improve our understanding. Some common concerns include:

❌ “AI isn’t actually thinking—it’s just predicting words.”
True! But that doesn’t mean it can’t help us think better. Just like a calculator doesn’t “understand” math but still helps us solve problems, AI can surface insights we might not have seen on our own.

❌ “Won’t AI just make people lazy?”
It depends on how we use it (I’m going to be addressing this in a future Podcast! 👀). If we rely on AI to think for us, that’s a problem. But if we use AI to expand how we think, it becomes a powerful tool.

❌ “AI will just flood the world with junk information.”
That’s a real risk! That’s why it’s important to be thoughtful about how we use AI. Cultural Optimists believe AI should be designed in ways that encourage depth and curiosity, not just endless content.

What This Means for Everyday Life

If AI really does have the potential to reshape how we understand the world, what does that mean for you?

• In Work: Instead of just automating tasks (not that this is bad, I think AI is great for certain task automation), AI could help you think more clearly, organize ideas, and improve decision-making.

• In Learning: AI could personalize education so that everyone learns in the way that works best for them.

• In Creativity: AI could spark new ideas, offering unexpected perspectives that lead to breakthroughs in art, music, and storytelling.

• In Conversations: AI could help people see different viewpoints, making discussions deeper and more productive.

In other words, AI isn’t just about making things easier. It’s about making things richer.

Looking Ahead: What’s Next?

Cultural Optimists see a future where AI is seamlessly integrated into our daily lives—not as something that replaces human intelligence, but as something that helps it grow. But they also recognize that this vision isn’t guaranteed. It depends on how we choose to use AI and whether we design it in ways that encourage exploration rather than just efficiency.

So here’s something to think about:

• Have you had an experience where AI helped you see something differently?

• Do you think AI will expand human understanding, or could it have the opposite effect?

• How would you want AI to help you think in your daily life?

The way AI develops will depend, in part, on how we interact with it. The question isn’t just what AI can do—it’s what we want it to do for us.

And that’s where things get really interesting.

A Plumb (AI automation) Workflow

You may recall that this time last year my team pivoted to build an AI automation platform (Plumb).

My co-founder Aaron built a meal planner inside of Plumb and it’s a game-changer.

It creates personalized meal plans based on your preferences, complete with recipes, grocery lists, and even emails the list to you. It’s just an example of the power of Plumb!

Check out the video and let me know if this kind of automation is interesting.

You can also see this video on Loom if you want to leave comments/emoji!

More Midjourney Image Exploration

As I mentioned previously, I’ve been experimenting with MidJourney to create these miniature dioramas of space—tiny, detailed worlds that feel like a mix of cosmic wonder and handcrafted art.

These are perfect as Notion covers or just a source of inspiration, these images capture the vastness of space in a playful, scaled-down way. Take a look and let your imagination wander!

I hope you learned something new or walked away with some inspiration.

I’d love to hear how my newsletters are sparking ideas or helping you see things from a different perspective.

Until next time, keep chasing wonder with curiosity! ✨🤘

Chase

This issue focuses on AI Concepts, unpacking the core ideas, frameworks, and principles that guide how we design and use AI.

One of the worst parts about being human is that it's in our nature to pick sides.

And nowhere have I seen it more clearly than when someone finds out I have a startup dedicated to helping people leverage AI.

The mood changes instantly depending on the "side" they've picked.

The doomers lean in, brow furrowed, face scowly, "AI is going to take all the jobs" or "AI is going to kill us all". They're certain we're headed towards catastrophe

The optimists light up, the evangelical energy building: "AI is going to solve all of humanity's problems!" This group is ready to merge with the machines.

And then we have the probers, the cautiously curious, asking "what do you think about AI and where it's going?" - trying to figure out which camp I belong to before revealing a position or that they don't have one at all.

What all three camps miss is that picking a side on AI is like picking a side on the future itself - it reduces something vast and complex into something small and manageable.

The real conversation isn't about choosing between artificial extremes. It's about mapping the vast territory that lies between them.

The reality is that most people working directly with AI technology don't fit neatly into any of these camps.

They see both extraordinary potential and legitimate concerns.

They recognize that different applications require different approaches.

They understand that the future isn't predetermined, but shaped by the choices we make today.

Beyond the Camps (A New Way to Think About AI)

This instinct to pick sides isn't just affecting casual conversations - it's actively shaping how AI is being developed, discussed, and deployed.

Watch any AI debate and you'll see it play out in real time.

One side argues we're racing toward digital superintelligence that will solve all of humanity's problems.

The other warns we're on the brink of losing control of increasingly powerful systems.

Meanwhile, businesses and developers caught in the middle are forced to choose between being labeled as reckless optimists or fearful luddites.

But what if there was a better way to think about our relationship with AI?

A framework that captures the nuance that most practitioners actually experience in their daily work?

After a lot of conversations with people building and implementing AI systems, patterns emerge.

Not three camps, but rather nine distinct positions - each with its own perspective on how AI should be developed and deployed.

This isn't just an academic exercise.

Understanding these positions - and more importantly, the underlying dimensions that shape them - gives us a much richer vocabulary for discussing AI's future.

It helps us move beyond simplistic debates about whether AI is "good" or "bad" and toward more meaningful conversations about how we want to shape its development.

In the next few newsletters, we'll explore each of these positions in detail.

But first, let's look at the two key dimensions that define this space: the pace of development and the nature of human-AI interaction.

The Two Dimensions That Matter

When you strip away the rhetoric and emotion, two fundamental questions shape how people think about AI:

• "How optimistic should we be about AI's development?"

• "Should we focus more on technical or cultural solutions?"

The first question - about outlook - reveals three distinct positions:

• The Optimists

• The Moderates

• The Cautious

The Optimists

The Optimists see AI as a powerful force for positive change.

They believe in humanity's ability to handle the challenges AI presents, whether through cultural adaptation, pragmatic innovation, or technical solutions.

In their view, the potential benefits of AI far outweigh the risks.

The Moderates

The Moderates take a balanced view, seeking evidence-based evaluation and measured progress.

They advocate for thoughtful development that considers multiple perspectives, whether through institutional reform, balanced assessment, or rigorous engineering approaches.

The Cautious

The Cautious voices emphasize the importance of getting things right over moving quickly. Whether focused on systemic risks, robust validation, or technical safety measures, they believe we need to deeply understand and address potential challenges before proceeding.

The second question - about approach - also reveals three distinct mindsets:

• Cultural/Systemic Thinkers

• Balanced Pragmatists

• Technical Focusers

The Cultural/Systemic Thinkers

The Cultural/Systemic Thinkers focus on broader societal implications and structures.

They emphasize how AI will interact with and reshape our social fabric, institutions, and ways of being.

They believe successful AI development requires addressing fundamental questions about society and human values.

The Balanced Pragmatists

The Balanced Pragmatists seek to bridge technical and cultural considerations.

They look for practical solutions that account for both human and technical factors, often serving as translators between different viewpoints.

The Technical Focusers

The Technical Focusers emphasize engineering solutions and concrete implementations.

They believe most challenges can be addressed through careful technical work, whether that means rapid innovation or rigorous safety measures.

When we combine these dimensions, we get nine distinct positions - each offering a different vision for how AI development should proceed.

Some see AI as a catalyst for human evolution, others as a technical challenge to be solved, and still others as a systemic risk requiring fundamental restructuring.

None of these positions is inherently right or wrong.

Each represents a thoughtful response to the unprecedented challenges and opportunities AI presents.

Understanding them helps us move beyond oversimplified debates and toward a more nuanced conversation about AI's future.

Where The Dimensions Meet

When these two dimensions intersect, they create nine distinct positions in the AI landscape.

Each captures a unique perspective on how we should approach AI development.

Let's explore each one.

The Cultural Optimists

They believe AI will fundamentally reshape human society - and that's a good thing.

"AI will help us understand ourselves better" isn't just a slogan for them; it's a deep conviction that our engagement with AI will lead to profound cultural and personal growth.

The Pragmatic Innovators

These thinkers balance enthusiasm with practical considerations.

Their motto - "Let's build thoughtfully but boldly" - captures their approach to keeping pace with AI advances while carefully considering their implications.

They're focused on real-world applications and evidence rather than pure speculation.

The Techno-Optimists

They maintain a strong belief that technical solutions exist for any challenges AI might present.

Their mantra is "Move fast, we can solve the problems," reflecting their confidence in humanity's ability to address issues as they arise.

The Reform Advocates

Their focus isn't just on the technology but on the systems that shape its development.

"Fix the systems, then advance" captures their belief that we need to address fundamental societal issues to develop AI responsibly.

The Balanced Moderates

"Let's consider all angles carefully" isn't just caution - it's a methodology.

These thinkers work to find consensus between different viewpoints, weighing evidence and considering multiple perspectives before drawing conclusions.

The Technical Pragmatists

This is an engineering-focused but measured approach.

Their motto - "Build carefully, test thoroughly" - reflects their commitment to rigorous testing and incremental progress.

They believe in the power of technical solutions but advocate for careful validation.

The Systems Risk Focus

These thinkers emphasize existential and systemic risks.

Their call to "rethink everything" stems from a deep concern about the fundamental structures shaping AI development.

They push us to question our basic assumptions about progress and development.

The Cautious Validators

Stuart Russell leads this group focusing on validation and verification.

Their approach - "Verify first, advance second" - emphasizes the importance of robust safety measures.

They believe we can develop AI successfully, but only with proper precautions and thorough validation.

The Technical Safety Advocates

These thinkers are focused on technical safety measures and mathematical rigor.

Their emphasis on "Solve alignment before scaling" reflects a deep commitment to addressing fundamental technical challenges before pursuing more advanced AI capabilities.

These positions aren't static camps but rather perspectives that often overlap and evolve.

Many thoughtful people find themselves drawing from multiple positions depending on the specific issue at hand.

A researcher might align with Technical Pragmatists on development methodology but share concerns with Systems Risk Focus thinkers about long-term implications.

Understanding these positions helps us move beyond simple pro/anti AI stances.

It reveals the rich landscape of thought shaping AI's development and gives us a better vocabulary for discussing its future.

Most importantly, it shows us that the most interesting conversations happen not when we pick sides, but when we understand the full spectrum of perspectives informing this crucial field.

Finding Your Position

Now comes the fun part - figuring out where you stand in this landscape.

You might find yourself drawn to multiple positions at once. That's not just okay - it's actually a good sign. It means you're thinking beyond the artificial extremes I mentioned earlier.

Try asking yourself these questions:

• How do you fundamentally feel about AI's impact on society? Are you excited about the possibilities, carefully weighing the evidence, or deeply concerned about the risks? This helps locate you on the optimistic-to-cautious dimension.

• When you think about addressing AI's challenges, do you find yourself drawn to cultural and systemic solutions, technical approaches, or a balance of both? This places you on the cultural-to-technical axis.

Your position might shift depending on the context.

You might align with Pragmatic Innovators when it comes to AI in education (enthusiastically but thoughtfully pushing for new learning tools), but share the Cautious Validators' perspective when it comes to autonomous systems.

You might even find yourself drawing from different positions on different aspects of the same issue.

Perhaps you share the Technical Safety advocates' concerns about AI alignment while appreciating the Cultural Optimists' vision of how AI could transform human understanding.

This framework isn't about picking a permanent camp - it's about understanding the landscape of possibilities and making conscious choices about where to stand on different issues.

Think of it as a map rather than a set of boxes.

Your position can move as you learn and as the technology evolves.

Most importantly, it gives us a better way to understand and discuss AI's future.

Instead of arguing about whether AI is "good" or "bad," we can have nuanced conversations about specific approaches and their implications.

We can better understand why someone might be a Systems Risk thinker on one issue but a Technical Pragmatist on another.

Next week, we'll dive deeper into each position, exploring real-world examples, the people who are most vocal in each position and the rationale behind different approaches.

For now, try mapping your own thinking - and maybe ask a few friends where they stand.

You might be surprised at the conversations that emerge when you move beyond the usual camps.

AI Generated Art

I’ve been experimenting with a style for AI generated images for content like this.

I really like the style that I used for the share image. Here are a few more that I’ve generated for similar content.

I’d love to hear what you think about this style (as opposed to the miniature diorama style below).

For reference, here are some of the miniature diorama images I was generating and using.

This issue explores Applied AI, showing how AI works in real projects and the lessons that come from building in the real world.

Imagine you’re asking a talented actor to perform a scene without giving them any context or character.

The result? A flat, generic performance. Now, imagine giving that same actor a detailed role—“You’re a courageous knight rallying your troops before a battle.”

Suddenly, the performance comes alive with depth and emotion.

The same is true for interacting with artificial intelligence.

When we engage AI with clear, defined roles, the results are richer, more accurate, and more aligned with our needs.

This simple shift, from action-based to actor-based prompts, can drastically enhance how you collaborate with AI.

What Are Actor-Based Prompts?

Traditional prompts often focus on issuing commands: “Summarize this document,” or “Explain quantum computing.”

These action-based prompts tell the AI what to do but lack the how.

This may be okay in cases where the tolerance for quality output is high, but when the tolerance for quality is low, you need a better way to increase the odds your quality will be high. This is where actor-based prompts come in handy.

Actor-based prompts assign a role or persona to the AI, such as, “Act as a journalist writing for teens and explain quantum computing.”

This subtle but powerful difference taps into the AI’s ability to simulate context and tone, producing outputs that feel more human-centered and relevant.

The Science Behind Role-Based Prompts

Actor-based prompts aren’t just a creative gimmick; they are rooted in cognitive and psychological principles.

Theory of Mind

When we assign a role to AI, we’re essentially asking it to adopt a perspective or framework for reasoning.

This mimics “theory of mind”—the ability to attribute thoughts, emotions, and intentions to others.

While AI doesn’t truly “understand”, role-based prompts guide it to simulate such reasoning, producing outputs that align more closely with human expectations.

Empathy and Role Adoption

Roles provide the AI with a lens through which to filter information.

For instance, when tasked to “act as a career coach,” the AI’s responses will inherently emphasize guidance and support.

This creates an illusion of empathy, making the interaction more engaging and accessible.

Empirical Evidence

Studies and user experiences consistently highlight the superiority of actor-based prompts:

• Accuracy and Relevance: Assigning a role reduces ambiguity, leading to more focused outputs.

• Creativity: Role-playing prompts unlock the AI’s potential for nuanced, imaginative responses.

• User Satisfaction: Users report greater alignment with their needs when interacting through defined roles.

Action-Based vs. Actor-Based Prompts: A Practical Comparison

To illustrate, let’s compare an action-based prompt to an actor-based one.

Action-Based Prompt Example

Let’s look at a very basic Action-Based Prompt for quantum computing:

Explain quantum computing.

The result of the Action-Based Prompt:

Quantum computing is an advanced computing paradigm that leverages the principles of quantum mechanics—the fundamental laws governing the behavior of very small particles like atoms and photons—to perform calculations. Here's an explanation broken down into accessible parts… (see the full response here)

Actor-Based Prompt Example

Let’s look at an Actor-Based Prompt for quantum computing:

You are a science journalist writing for teens. Explain quantum computing in an engaging way with examples.

The result of the Actor-Based Prompt:

Imagine you’re in a massive library with trillions of books, and you’re tasked with finding one specific book. If you’re a regular computer, you’d start at one end, checking book by book until you (finally!) find the right one. That’s a bit like how computers work now—powerful, but slow when tackling enormous problems. Now imagine you have a magical power that lets you open all the books at the same time. That’s the magic of quantum computing. It’s not magic, of course, but it might feel like it when you see what it can do… (see the full response here)

The actor-based prompt produces a more tailored and engaging response, highlighting the importance of context and framing.

Why It Works

Human-Like Interactions

By assigning roles, we prompt AI to simulate attributes of that persona, leading to outputs that feel more intentional and empathetic.

Cognitive Load Reduction

Clear role assignments reduce the need for users to refine or correct responses, saving time and effort.

Accessibility for Non-Technical Users

Actor-based prompts simplify interactions by making the AI feel less like a complex machine and more like a collaborative partner.

Real-World Applications

Actor-based prompts are reshaping how AI is used across various domains.

By assigning specific roles, users can unlock new levels of utility and creativity in their interactions with AI tools.

For example, a healthcare professional might prompt AI to “act as a patient advocate” to draft empathetic communication for sensitive diagnoses, while a project manager could request AI to “act as a productivity consultant” to suggest workflow improvements tailored to their team.

These tailored prompts don’t just make AI more effective; they also make it feel more collaborative and personalized, bridging the gap between technical complexity and everyday usability.

For Non-Technical Users

• Small Business Owners: Prompt AI to “act as a marketing consultant” to get tailored advice for campaigns.

• Educators: Ask AI to “act as a historian” to generate engaging lesson plans.

For Technical Users

• Engineers: Use AI as a “code reviewer” to debug efficiently.

• Data Analysts: Prompt AI to “act as a storytelling expert” to translate raw data into compelling narratives.

Cross-Disciplinary Examples

• A writer using AI as a “poetry editor” for refining verses.

• A student asking AI to “act as a personal tutor” for breaking down complex topics.

How to Write Effective Actor-Based Prompts

When I’m thinking about a prompt, I use this order of operations for writing an actor-based prompt for what I need. I also use an AI meta prompt to help me generate these prompts.

Here’s the meta prompt you can use to create your own actor-based prompts and unlock the full potential of AI:

Meta Prompt: “Help me craft an actor-based prompt by following these steps: First, identify the specific role I want the AI to take (e.g., a teacher, a coach, a marketer). Second, define the task clearly, specifying the context and the audience for the output. Third, describe the tone or style I’d like the AI to use (e.g., conversational, formal, playful). Lastly, suggest ways I can refine the prompt further to improve the output.”

Using this structure ensures your prompts are clear, focused, and tailored to your needs.

1. Define the Role: Be specific about the persona (e.g., “You are a financial advisor with expertise in retirement planning.”).

2. Set the Scene: Provide context for the task (e.g., “Explain this to someone with no financial background.”).

3. Specify Tone and Style: Tailor the response further (e.g., “Be conversational and engaging.”).

4. Iterate and Refine: If the output isn’t perfect, build on it (e.g., “Make this sound more like a TED Talk.”).

The Bigger Picture: Human-Centered AI

Actor-based prompts align with a broader vision of making AI more accessible and human-centered. By framing interactions around roles, we:

• Democratize AI Access: Simplify complex tools for non-technical users.

• Foster Curiosity and Creativity: Encourage playful experimentation.

• Build Confidence: Empower users to engage deeply with AI, unlocking its full potential.

Conclusion: Bridging the Gap

Actor-based prompts transform AI from a functional tool into a collaborative partner.

By inviting AI to “play a role,” we open doors to richer, more engaging interactions.

Whether you’re a small business owner, a teacher, or a software engineer, this approach makes AI feel more intuitive, empowering, and accessible.

If you’re new to AI and feel unsure about how to get started, I’d be happy to help! Reach out to me for an introduction to AI tools and tips to make your first interactions more comfortable and rewarding.

Challenge: The next time you interact with AI, try assigning it a role. Notice how the responses change, and share your experiences. Together, we can continue to explore the exciting possibilities of human-centered AI.

Key Takeaways

• Actor-based prompts outperform action-based ones by providing clarity and context.

• Assigning roles improves relevance, creativity, and user satisfaction.

• Start small: Experiment with a specific role and refine as needed.

By transforming technical barriers into bridges of understanding, actor-based prompts invite us all to engage with AI in a way that feels both powerful and personal.

✨🤘Stay curious!

"The future is not something we enter. The future is something we create." - Leonard I. Sweet

When scientists first peered through microscopes, they didn't just see more clearly—they discovered entirely new worlds.

We’re in the middle of a similar moment (maybe even more impactful) with artificial intelligence (AI)?

Just like microscopes expanded our vision into the microscopic, AI is extending our thinking into uncharted territories.

It's not just a tool to make tasks easier; it's a collaborator that helps us explore new possibilities and achieve things we couldn't on our own.

Rethinking AI: It's Not About Replacement

You've probably heard people say AI will replace jobs or even creativity.

But what if we stopped seeing AI as a replacement and started viewing it as a lever?

A lever doesn't do the work for you—it amplifies the effort you're already putting in. AI is the same.

It doesn't replace your creativity, intuition, or judgment. Instead, it strengthens these qualities and helps you go further, faster.

Think of AI as a brainstorming partner, research assistant, or creative sounding board—always ready to enhance your ideas, not take them over.

Three Simple Ways to Use AI in 2025

Here's how you can think about AI as your personal lever:

1. Automate to Elevate

Smart automation isn't about replacing human work—it's about freeing you to focus on what matters most. Use AI to handle repetitive tasks, giving you more time and mental space for creative thinking, strategic planning, and meaningful interactions.

Try this: Identify your most time-consuming routine tasks (like email sorting, data entry, or basic research). Set up AI automation for these, then notice how you use the time you've gained. What becomes possible when you're not caught up in the routine? (Check out Plumb, the workflow automation platform my team is building, if you’re looking for a way to do this!)

2. Collaborate, Don't Delegate

The best results happen when you work with AI, not just hand tasks off to it. Treat AI like a collaborator that offers fresh perspectives, challenges your thinking, and sparks creativity. Your expertise combined with AI's insights creates something neither could achieve alone.

Try this: Start a dialogue with AI about a challenge you're facing. Ask questions, challenge its suggestions, and use its responses as springboards for your own thinking.

3. Enhance, Don't Replace

AI is at its most exciting when it enhances your capabilities in ways you didn't think possible. Think of how a telescope lets you see galaxies or a microscope reveals hidden cells—AI does the same for your thinking. It helps you spot patterns in data, connect ideas, or explore creative directions you wouldn't have seen otherwise.

Try this: Use AI to analyze a dataset or piece of writing from multiple angles. What patterns or connections does it reveal that you might have missed?

Practical Ways to Explore AI Right Now

For Creative Projects

• Use AI as a creative sparring partner: Share your ideas and ask it to help you explore different directions or overcome creative blocks.

• Create mood boards or inspiration collections by describing what you're looking for and letting AI suggest unexpected combinations.

For Research and Analysis

• Ask AI to summarize complex topics from multiple perspectives, helping you build a more comprehensive understanding.

• Use it to generate questions you haven't thought of, pushing your research in new directions.

For Learning New Skills

• Create personalized learning paths that adapt to your pace and style.

• Generate practice scenarios that challenge you at just the right level.

Your Unique Role in the Collaboration

Remember, a lever only works if someone knows how to use it. AI is no different—it needs your creativity, judgment, and intuition to unlock its potential.

When working with AI, ask yourself:

• What unique perspective can I bring to this collaboration?

• How can I use AI to build on my strengths rather than replace them?

• Where could AI help me think bigger or explore new possibilities?

The magic happens when you combine what makes you human with what makes AI powerful.

Your Next Step: Start Small and Stay Curious

Ready to give it a try? Here's your action plan:

1. Pick one task where you could use some extra insight or inspiration

2. Experiment with using an AI tool as a collaborator

3. Reflect on how it changes your process—what works, what doesn't, and where it sparks new ideas

The future of AI isn't about humans versus machines.

It's about humans and AI teaming up to create something greater than either could alone.

By approaching AI as a partner, you can begin to amplify your strengths and explore exciting new possibilities.

Did anything in this issue stand out or give you energy? If so, I’d love to hear from you about it so I know what to focus on sharing in the future!