Uncrowded Strategies #021: I built a tool that makes three AIs fight, then keeps the winner.

Hey, it's Gabriel.

These are my private notes this week on building wealth, raising my family, and breaking generational patterns.

This week, I’m sharing with you a tool that pits Claude, Gemini, and ChatGPT against each other, the calmest delivery ward the nurses had ever seen, why I'm going to cash before the midterm election, and the sleep rule my WHOOP data confessed.

I built a tool that pits three AIs against each other and then hands me the winner. I'm giving it away free this week.

Here's the trap almost everyone is stuck in. You ask one AI a question. It answers like it is certain. And you have no way to know if it is wrong, because there is nothing sitting next to it to check it against. One model is one opinion. A confident wrong answer is the most expensive kind of wrong there is.

Most people try to fix this by getting better at prompting. I went the other way. I stopped relying on one model at all.

I call it LLM Fusion. It runs in two modes.

Council mode. Your question goes to Claude, Gemini, and ChatGPT simultaneously. Each one answers on its own, blind, never seeing the others. Then a judge model reads all three with the names stripped off and gives you the strongest combined verdict. It's like walking out of one doctor's office and quietly getting the same scan read by two more, then a senior consultant who never saw who said what.

Andrej Karpathy first demoed a version of this idea, so credit where it's due. The catch in his version is that it runs on API keys, and that bill climbs fast. Mine runs on the subscriptions you already pay for. If you have Claude, Gemini, and ChatGPT, you run the entire council on tokens you already bought. Three $20-a-month plans most of us already have open in our browser. Nothing extra to spend.

Build mode. This is the part I'm most excited about. You give one instruction. Every agent writes out, step by step, exactly how it would do the job. The judge reads every plan, picks the best one, and then it actually executes and builds the thing. Not three opinions for you to referee. One finished result, chosen from the best thinking of all of them.

The "Uncrowded" Move: On your next real decision, don't accept the first answer from the first AI. Make them compete. The disagreement between models is exactly where the truth has been hiding. Grab LLM Fusion free here

A quick word on why I just give this away. Inside my Two-Hour CEO community, this is a normal week. I show, step by step, how to set up these tools and point them at building something useful: an agent that takes over a workflow you hate, or an automation you can build and sell to another business. Because here is the quiet opening that most people walk past. So many companies need this and have nobody to build it. You don't need to be a developer to be that person. You're not learning to code. You're learning which tool to point at the problem. I'm not an engineer teaching theory. I'm a CEO whose own agents run my real company, and I hand you the same ones. You can join our Two-Hour CEO Skool here

Last week, my third child was born, and the delivery ward couldn't explain what they were looking at.

That's why this letter went quiet for a week. Worth it.

No epidural. No pain medication. We walked into the labor ward, sat down, and the nurses could not believe how calm Zoey was. Some of them have done this work for thirty, even forty years. They told us, plainly, that they had never seen anyone this calm in active labor.

This is the third time we've witnessed it. For us, it comes down to faith. We believe the pain of childbirth was part of a curse, and that Christ redeemed us from that curse. So a mother can claim a different promise, because children are a gift from God, and a gift is not supposed to arrive wrapped in pain.

I know how that sounds. The AI guy is talking about lifting a curse in a maternity ward. Believe whatever you want about the why. I'm only telling you what three nurses, with a combined 100 years of experience between them, said they had never seen before.

What pain in your life have you simply accepted as the price, without once asking whether it was ever yours to carry?

In this family, we don't inherit the curse. We claim the promise.

The "Uncrowded" Move: The biggest shifts rarely come from working harder inside a rule everyone accepts. They come from asking whether the rule was ever true. If you want to understand where our conviction on this comes from, the book that shaped ours is Supernatural Childbirth by Jackie Mize. Read it whether you're expecting or not.

I'm trimming profit and moving a little to cash, right as the most hyped IPO in history hits the tape.

The SpaceX IPO was wild to watch. It was priced as the largest IPO on record, raising around $75 billion, then popped close to 30 percent on its Nasdaq debut. All in the same week, the Fed leaned hawkish on rates. Quite a thing to sit and watch.

I'll be honest, I didn't do much in my own portfolio this week. I was busy holding a newborn. But here is what I'm weighing as we head deeper into the year.

Watching: the calendar. 2026 is a US midterm election year, and history has a rhythm worth respecting. Midterm years tend to be the rocky, muted ones, with weakness through the summer and a risk of correction building into the vote. Then markets have historically rebounded hard afterward, with the following year one of the strongest in the entire four-year cycle.

The move: trim a bit of profit and hold maybe 10-15% in cash heading into the midterms. Dry powder, so a correction becomes an opportunity instead of a gut punch.

The honest part: the AI boom is unlike anything in past cycles, so the old pattern may not play out the same way this time. I'm not calling a top. I'm staying a little liquid and staying ready.

The Uncrowded Take: You don't have to predict the market to respect its rhythm. Going modestly to cash before a historically rocky stretch isn't fear. It's keeping ammunition for the sale. (None of this is financial advice. It's just what I'm doing with my own money.)

I exported four years of WHOOP data, fed it to Opus, and it found a rule about my sleep I'd never have caught on my own.

Most people don't know you can even do this. Open the WHOOP app, go to Settings, tap Export Data, and your entire history is saved in a single file.

I dropped mine into Opus and ran a deep analysis in Claude. I used Fable for part of it too, before it got pulled, and I'm hoping it comes back soon.

The insights were sharper than anything the app dashboard ever showed me. The one that stopped me cold: every single time I sleep later than 1 am, my recovery tanks the next day. It doesn't matter how long I'm in bed. A full eight hours after a 1 am bedtime, and my recovery is still in the gutter. The timing matters more than the duration.

That's a pattern you'll never catch glancing at a daily score. The app shows you today's weather. The export shows you the climate.

This is where health is heading. Your blood tests, your sleep tracker, your food logs, all of it has been sitting there as dead data. We finally have something that can actually read the whole pile at once.

The "Uncrowded" Move: Export your WHOOP (Settings, then Export Data), drop the file into Opus 4.8, and ask it to run a deep analysis of your patterns. The one rule it surfaces will do more for you than the score you anxiously check every morning.

As promised, my private notes. No fluff.

See you next week,

Gabriel Judah

P.S. Over 200 of us inside Two Hour CEO are building these agents in plain files we own, the ones that quietly take over the work we used to dread. You don't need to be a developer. You need to know which tool to point at the problem, and that's what I teach, step by step, every week. Grab a seat here.