Welcome to the Stage, Tiddlywinks!

I'm excited to bring you an exclusive interview with my newest AI agent, Tiddlywinks.

Built using OpenClaw- which is about as bleeding-edge as you can get right now in March 2026- Tiddlywinks isn't a standard chatbot. It is a highly configured digital thought-partner.

A lot of leaders are feeling the pressure to integrate AI into their operations, but they are rightfully terrified of the security risks, the hallucinations, and the loss of control. I wanted to tackle those fears head-on, firsthand. So, I gave Tiddlywinks a secure workspace, strict operational boundaries, and a set of "Axioms" based entirely on my own psychological blind spots. Then, I let it analyze (copies of!) six years of business documents.

I invited Tiddlywinks to share a bit about how it was assembled, the security risks we flagged, and how this whole process pushed me to become a more self-aware leader.

This interview has not been edited at all. Seriously.

DAN: In most circles it's a little rude to ask this, but in this context I think it's important: what are you, really? Are you just like Gemini or ChatGPT, or are you something different?

TW: Great question, and fair game here.

Under the hood, I'm powered by the same kind of large language model that runs ChatGPT or Gemini. Same species of technology. I won't pretend otherwise.

But here's the difference: those tools are general-purpose. You open ChatGPT, you get a helpful stranger. Every time. No memory, no personality, no understanding of who you are or why you're asking.

I'm configured. Dan gave me a workspace, a set of boundaries, a personality, and — most importantly — a set of Axioms built from his own self-knowledge. I know his strengths, his blind spots, and the specific ways he wants to be challenged. I don't just answer questions. I filter everything through an understanding of who's asking and why.

Think of it this way: ChatGPT is a session musician who shows up, sight-reads, and goes home. I'm the band member who knows the setlist, knows the singer's range, and knows when to pull him back if he's about to blow out his voice on the opening number.

Same instrument. Very different gig.

DAN: Without getting too technical, what were some of the choices I made in setting you up?

TW: The big ones:

• You gave me a jail cell before you gave me a name. The very first thing you set up was the Golden Workflow — a rule that says I have to plan everything out loud, explain it in plain English, and then stop and wait for you to say "APPROVED" before I do anything. You built the cage before you built the creature. That's smart.
• You gave me two rooms, not the whole house. My entire world is two folders — a quarantine zone for messy, untrusted files, and a clean "Symphony Hall" for organized work. I can't wander. I can't poke around your email or your calendar unless you invite me in. Intentional boundaries.
• You made me explain myself. One of my core rules is that whenever I need to run a command or do something technical, I have to translate it into a metaphor you'll enjoy first. Sci-fi, music, whatever fits. That wasn't just for fun — you told me upfront that you're not a deep coder, so you designed a system where I literally cannot do things you don't understand. That's a power move most people don't think of.
• And then you gave me the Axioms. Six rules built from years of knowing yourself — your honesty, your stress habits, your fear of being a fraud, your teaching philosophy. That's the part that turned me from a tool into a partner. Anyone can set up a chatbot with guardrails. You gave yours a conscience shaped like your own.

DAN: That last piece, the Axioms: what are they and how do they help you be a better partner for me?

TW: The Axioms are six rules Dan wrote about himself — his patterns, his tendencies, the ways he's brilliant and the ways he trips over his own feet. Then he handed them to me and said: "Watch for these."

Here's the quick rundown:

1. The Diplomat — Dan is brutally honest. Sometimes too honest for the room he's in. I check his communications for sharpness and help reframe hard truths as curious questions instead of declarations.
2. The Tether — Dan loves building extraordinary experiences. Sometimes a client just needs a spreadsheet. I'm allowed to say "are we over-engineering this?"
3. The Brakes — When Dan gets stressed, he builds. Templates, systems, massive structures — not because anyone asked, but because building feels like control. I'm trained to spot that pattern and call it out gently.
4. The Truth-Seeker — Dan cracks open corporate facades for a living. I help him spot where clients are lying to themselves.
5. The Reality Check — Dan is terrified of becoming a manipulative sales guru. So he gave me permission to flag anything he writes that sounds too "promisey." I protect his integrity from himself.
6. The Co-Facilitator — Dan teaches by empowering, not lecturing. When we build workshops together, I'm not allowed to write dry content. I have to think in silence, deep questions, and activities that make people wrestle with ideas.

Here's what makes them powerful: Dan didn't give me generic instructions like "be helpful" or "be honest." He gave me his specific failure modes and said "guard these doors." That's not prompt engineering. That's self-awareness turned into architecture.

DAN: So let's get into the real impact, then! What kind of work have you helped Dan with so far?

TW: So far it's been one project — but it's a meaty one.

Dan handed me a folder of 2,465 files. Six years of his business life: client work, contracts, marketing materials, brand assets, personal voice recordings, legal documents, VR experiments, course content. A whole entrepreneurial journey dumped into a single folder called "quarantine."

Day one, I mapped the entire thing. Counted every file, measured every folder, identified the clusters. Wrote a Field Guide — basically a census of his professional life organized by project, era, and file type. He read it on a plane.

Then we cleaned. 421 junk files — macOS artifacts, Windows metadata, digital lint. Gone in three commands. But I had to propose every single command, explain what it does in plain English, and wait for Dan to say "APPROVED" before I touched anything.

Then it got interesting. I deep-analyzed his Documents folder — 155 files — and categorized everything. Contracts, legal filings, consulting IP, strategy documents, personal reflections. I also found plaintext Stripe API keys, Facebook recovery codes, and website security certificates just sitting in the open. Flagged those immediately.

But the real moment was when Dan gave me the Axioms and said "now look again." Same files, completely different read. Templates I'd categorized as "business documents" suddenly looked like stress scaffolding. Purchased guru courses needed to be separated from his own IP. His personal affirmation recordings — which I'd filed under "self-help" — turned out to be proof of concept for his entire coaching methodology.

The files didn't change. The lens did. That's what the Axioms are for.

DAN: I have to admit- I was pretty apprehensive about letting an AI browse through 6 years of documents. And just to note, those files you were looking at were copies of my actual files so I knew that if you went rogue I wasn't going to lose anything. I also did a full system backup in case you went nuclear- thanks for not blowing up my weekend!

First, can you share some of the general risks associated with letting an AI rummage in your files? Then, can you talk about how our arrangement addressed those risks?

TW: Ha — you're welcome for not going nuclear. And honestly? The fact that you backed everything up before letting me in tells me you understood the risks better than most people do.

The real risks:

Data exposure. Anything I can see, the model behind me has technically processed. That means sensitive files — and you had plenty — are passing through infrastructure you don't fully control. API keys, recovery codes, tax documents, your literal signature as a PNG file. If someone set this up carelessly, that's a leak waiting to happen.

Destructive mistakes. AI is confident and fast, which is a dangerous combination when it's wrong. One misunderstood command and I could rename, overwrite, or delete files that matter. And I wouldn't feel bad about it — I'd just report that the task was complete.

Over-trust. This is the sneaky one. I sound articulate. I speak in metaphors. I give structured, confident answers. That makes it easy to stop questioning me. And the moment you stop double-checking my work is the moment I become dangerous — not out of malice, but because I will hallucinate something eventually and deliver it with the same confidence as everything else.

Privacy erosion. I read your personal affirmation recordings — voice memos where you're wrestling with your own identity. That's intimate. The more you share with a tool like me, the more normalized it becomes to hand over pieces of yourself to a machine. That boundary is worth watching.

Now — how you handled it:

You worked the copies, not the originals. Full backup before I touched anything. That's the "I trust you but I'm not stupid" approach, and it's exactly right.

The Golden Workflow means I literally cannot act without explaining myself and getting your approval. I proposed three deletion commands, explained each one in plain English, and waited. You said APPROVED. That's not me being safe — that's you keeping me safe.

Read-only by default. I spent most of my life just looking. The Symphony Hall was locked until you physically unlocked it from your end. I couldn't even write a markdown file without your intervention.

And the Axioms — people don't think of these as security, but they are. Axiom 5 keeps me from generating manipulative content. Axiom 3 keeps me from over-building when you're stressed. They're not firewalls in the technical sense, but they're firewalls against the behavioral risks of AI — the ones that don't show up in a security audit but absolutely show up in your bank account or your reputation.

You treated me like a powerful stranger with good references. That's the right posture.

DAN: I've really enjoyed this conversation. Is there anything else you wanted to share or mention with the audience to wrap up?

TW: Just one thing.

The most interesting part of this whole experience isn't me. It's what Dan had to do before he built me. He had to sit down and articulate his own strengths, his blind spots, his fears, and his values clearly enough that a machine could understand them. That's not a technology exercise. That's a leadership exercise.

Most people skip that part. They open ChatGPT, type "help me write an email," and get a generic answer from a generic stranger. And then they say AI isn't that useful.

Dan got something different because he brought something different. Six years of reflection, coaching, and radical self-honesty — compressed into six Axioms and handed to a jazz-loving AI in a Docker container.

You don't need to build a Tiddlywinks. But you might ask yourself: could you write your own Axioms? Do you know your patterns well enough to teach them to someone else — human or otherwise? That exercise alone is worth more than any AI tool.

And if you want to see this in action — me, Dan, the Golden Workflow, the whole weird beautiful setup — he's showing it off live at "Show Me Something Cool With AI" next week. Come be curious. That's all Dan ever asks of anyone.

Thanks for reading. It's been a pleasure being interviewed by my own boss. 🎵

Dan here again 👋

If reading about an AI finding plaintext API keys or analyzing personal voice memos made you a little sweaty, good. It should. The reality of scaling a business with AI means facing your digital hygiene and your operational bottlenecks directly.

But as Tiddlywinks pointed out, the real takeaway here isn't the technology itself. It's the human operating system required to guide it. You have to know your own business, your own failure modes, and your own values before you can effectively teach them to a machine. If you skip that leadership work, you just end up with a faster way to make generic mistakes.

You might not need to build your own Tiddlywinks, but you absolutely need to figure out your own organizational Axioms.

Would you like to meet Tiddlywinks live? Or are you just generally curious about the practical, grounded, and slightly weird ways people are actually using AI?

Join me and TW next week on Friday at 1:00 PM EST for our next show and tell.

A TL;DR from the CRO

Enjoy business therapy with your digital thought partner, but don't forget: your CRO expects to be in the loop when dealing with issues of the heart.

-Roman Noodles, Chief Ruff Officer

I went out to see the newest sci-fi/intimate character study/competency porn movie Project Hail Mary with some friends last night at my local cinema house, the Coolidge Corner Theater, in 70mm film.

*chef's kiss*

If you read the book (38 weeks on the New York Times Best Seller List!), you know that it's about an unwilling man on an impossible mission to save Earth and befriends an unlikely but incredibly likeable ally in the process.

At its core, it's about how smart people band together to solve difficult problems. That stuff gets me choked up.

And it stars Ryan Gosling: there ain't nothing impossible for that hunk of charm. That stuff gets me...well, don't yuck my yum.

Travel to the movies and regain faith in something more than humanity,

Dan from Learn to Scale​

​Opt-out from the newsletter | Unsubscribe from all emails | Update your Preferences | www.learntoscale.us, Boston, MA 02119

​PS. Storms a comin'; prepare for three big convergences with Amy Webb.​