Uncrowded Strategies 020: I bought a $3,000 cooker and called it a health strategy

Hey, it's Gabriel.

Here’s another issue of Uncrowded Strategies. These are my private notes this week on my experiment with AI, building wealth, raising my family, and breaking generational patterns

I told my AI to review my own project. It spawned dozens of agents and found problems I'd missed for months.

Claude Opus 4.8 dropped last Thursday, May 28. I've had about a week with it.

I'll be honest, the benchmarks didn't make me sit up. Anthropic's own words for the gains are "modest but tangible." If you came for a headline leap, it isn't here.

The thing that actually changed how I work is buried inside Claude Code, and it's called dynamic workflows. Still in research preview, so early days. But one keyword inside it has already paid for itself: ultracode.

Here's what I did. I took a project I've been building for months, my personal dashboard, and typed one line: use ultracode to do a comprehensive review and tell me where this drifts from the plan.

Then I watched it fan the job out across dozens of agents.

Each one read a small slice. One read my data. One read the FAQ. One checked another corner. Then a final agent stitched everything together and handed me a list of weak spots I'd never have caught myself, because I was too close to it. The kind of stuff you stop seeing after staring at the same project for weeks.

The clever bit is why it doesn't choke. Normally you hand a model a big project and it drowns. Too much in the context window, and it starts making things up to fill the gaps. Dynamic workflows keeps each agent's reading small, holds the intermediate work outside the main context, and only the final answer comes back. Think of it like a newsroom. Twenty junior reporters each chase one lead, the editor reads only the filed stories, not every interview tape. The editor's desk never gets buried.

It even cross-checks itself. Claims that don't survive the review get filtered out before they reach you. Less likely to hallucinate, because the work got marked by another agent before it landed on my desk.

I've been using it to crunch a mountain of numbers for that dashboard, the sort of job that used to leave me babysitting one overloaded chat. Now I ask once and go make coffee.

The real win is that you quietly stopped being the bottleneck.

The "Uncrowded" Move: Take the one project you're too close to, open Claude Code, and tell it to ultracode a review against your own plan. The blind spots it surfaces are the ones costing you the most, precisely because you stopped seeing them.

I’ve been using Beehiiv to run my newsletter, and they will be announcing something big in July.

We wrote a constitution for our family, and the kids will get their voice heard before they can spell the word "constitution."

This week is my eighth wedding anniversary, so I did the romantic thing and took a day off with Zoey to write legislation.

I'm not joking. We sat down and started drafting what we call our family constitution.

Last week I told you, Zoey, and I do a yearly review of our marriage, almost like an AGM. This year, that same energy spilled into something bigger: our vision, our values, what the Judah family stands for, what we say, and what we don't say. All of it written down in one place, we can come back to.

Here's why I bothered. You don't break a generational pattern by feeling strongly about it. The pattern your parents handed you was never written down anywhere either, and look how reliably it still got passed on. A pattern in your head is like a handshake deal with your future. Easy to make, impossible to enforce. So if I want my kids to inherit something different, it has to be on paper, or the old defaults just win on autopilot.

The part I'm most excited about is a monthly family council we're building in. My kids are still very young, so let's be honest, for now, the council is Zoey and me making every decision while two small people lobby aggressively for more snacks. But this makes sure their voices are heard, they can give feedback on whatever they want.

But as they grow and take on more responsibility, they earn the chance to actually vote on what we do as a family.

I hope this gives them ownership, and a kid with ownership is proud to carry the name.

If your family is the most important thing you'll ever build, where is the document that says what it's actually for?

And it's not carved in stone. A constitution can be amended. As we grow, it grows. We can rewrite a clause next year if the family needs it to.

The "Uncrowded" Move: Write your family one page. Not a legal document. Just your values, your vision, and what you stand for, in one place you can come back to and amend as the years go. Write the pattern down, or the pattern keeps writing itself.

My agentic-bottleneck bets are paying off, and I'm still buying the names the crowd is too scared to touch.

For months, I've been writing about the OpenClaw moment, which is my name for the point where AI agents stopped being a toy and started flooding the internet with traffic. When an agent does your work, it touches thousands of endpoints where a human touches five. All of that traffic has to pass through something. The bottleneck is cloud, cybersecurity, and compute.

I named a few names. Cloudflare (ticker NET) was one of them.

NET is up around 40-50% from when I bought near the post-earnings dip. The whole market panicked when they announced a layoff, the stock fell by about a quarter in a day, and I bought into that fear. Their own CEO now talks about processing hundreds of billions of agentic requests a month. The thesis is playing out in their own numbers.

I exited UiPath (ticker PATH) at roughly break-even. Nothing wrong with the business. It actually just posted its first-ever GAAP profit. It simply went sideways while better setups ran, so I rotated. Why hold a stock at flat when the same dollar can work harder somewhere else?

That "somewhere" is ServiceNow (ticker: NOW).

Here's the bottleneck it sits on. An AI agent can think, but the hard part is getting that decision executed across a company's IT, HR, and finance systems that don't talk to each other. ServiceNow is the workflow layer that already runs that plumbing, so when companies switch on AI agents, it becomes where the work actually gets done. Cloudflare unclogs the network. ServiceNow unclogs the work inside the building.

NOW is down close to 50 percent from its all-time high. Not because it broke. The market got scared of a "SaaS apocalypse" and re-rated every enterprise software name on vibes. Meanwhile, the actual company is growing revenue over 20 percent a year, profitable, and is now wiring agentic AI into the core product. Strong company, fear-driven price.

That is the whole game in one trade. The crowd sold the story, and I bought the fundamentals.

The Uncrowded Take: Everyone piles into the stock the crowd is excited about. The real edge is the good company the crowd is scared of. ServiceNow at half off its high, bought into the fear, beats chasing whatever is already running. (And none of this is financial advice. It is just what I'm doing with my own money.)

I'm building an AI that decides what I eat, then tells my new $3K cooker exactly what to make.

A couple of weeks ago, I told you I'd started wiring Hermes into my body, not just my business. Whoop on the wrist, sleep and recovery, and strain flowing in every morning. That was the read layer. This week, I started building the act layer.

I bought a Thermomix TM7.

I'll be honest, a friend convinced me, and for about a day, I felt like the guy who buys a $3,000 cooker and calls it a health strategy. Then I saw what it unlocks.

The TM7 is Vorwerk's current model, out since early 2025, and it runs an app called Cookidoo. The new bit: you tap a recipe on your phone, and it lands straight on the machine. So the cooker doesn't just cook. It can be told what to cook, remotely, by something that isn't me.

That something is the AI agent I'm building. Call it BodyOps Agent.

Here's the pipe I'm stitching together. Whoop and my other health data go into Hermes. BodyOps reads it, decides what my body actually needs for the week: calories, macros, the recovery picture, and turns that into the meals I should eat. Then it sends those meals to the TM7 to cook. To be clear, that last hop isn't one magic button. It's an agent plus a third-party recipe importer plus Cookidoo's handover feature, taped together by me.

The missing piece was taste. A nutrition agent will happily tell you to eat the same sad chicken and broccoli for the rest of your life. So I plugged in an MCP called Epicure.

Epicure handles flavor pairing. It's early, so I'm experimenting, not declaring victory. It does one job and does it well: flavor. Which foods pair, what to substitute, and how to point a dish toward Southeast Asian or Mediterranean?

So the division of labor is clean. BodyOps decides what's good for me. Epicure makes what's good for me taste good. The TM7 cooks it.

None of this fully runs yet. I'm at the wiring stage, importer here, MCP there, and an agent learning my body in the middle. But the shape is now obvious to me: these agents are no longer software that runs a business. They're starting to run a life.

I'll share the data once it's had a few weeks to actually cook.

The "Uncrowded" Move: The future isn't asking your AI to plan your meals. It wires to the appliance that makes them, and does the planning, the taste, and the cooking.

See you next week,

Gabriel Judah

P.S. We still have 21 seats for Two-Hour CEO Skool. Over 200 members now are learning how to build an AI second brain, and automating their tasks and even boring jobs with AI Agents, buying back time to do the things they love. Grab a seat here.