Private notes, made public. · 4th Sep, 2026

Hey, it’s Gabriel.

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

This week
⚡ two weeks stuck on one problem, and the model that found what I could not see
❤️ the sentence that reframed what it means to be needed
💰 the market called it a SaaS apocalypse and dragged down the wrong companies
🧬 what a mouse study taught me about running four AI agents at once

Here is what I was stuck on, because the problem matters more than the tool.

I have spent a long time building what I call a second brain. Every note, every decision, every transcript, every lesson, searchable by my agents. It works. And this month I hit its ceiling, which is that it is *mine*. It has my therapy notes in it. It has my financial positions. It has conversations about my kids. I cannot hand that to my team, and a company that cannot share its memory is a company where every new person starts from zero.

So the real build is not one brain. It is two, with a door between them.

A personal brain that stays private. A company brain my team can actually work in, where different people can add to it without overwriting each other, where permissions are real, and where the two brains can pass things to each other without leaking what should never cross.

That is a genuinely hard design problem, and I was two weeks into it going in circles.

Now the honest part about the tools.

Fable 5 was great but it felt nerfed after the re-release following US export control ban.

So when Fable 5.1 landed on Tuesday, I was not expecting much.

I pointed it at the whole architecture. Two weeks of design decisions, the permission model, the sharing rules, all of it, and asked it to review the process.

It found the things I could not see. Not small things. Structural ones, in the part of the design I was most confident about. Then it kept going and flagged three problems I had never thought to ask about, which is the part that actually got my attention. I can ask a model to check my work. I cannot ask it to check the questions I did not know to ask.

That is the difference between a faster tool and a better thinker. Speed helps with known unknowns. This helped with unknown unknowns.

Most people would call it excessive to pay premium rates for a model to read a design document. And they would be right, if you used it for everything.

So here is the mechanic, and it is the whole reason this is affordable.

You do not hire the structural engineer to lay the bricks.

Fable 5.1 is expensive, so it never touches execution. It plans, it reviews, it judges. Then the actual building goes to cheaper models: Claude Opus 5, or GPT-5.6 if you already pay for ChatGPT, which most people do. The engineer draws the load paths. The crew pours the concrete. Nobody sends the engineer up a ladder with a trowel.

I built several large pieces of my knowledge work this way in the last three days and did not come close to burning through my weekly limits.

Which of your problems are you solving with a fast tool, when what you actually need is a slow one that thinks?

That is Arthur Brooks, and I have not been able to put the sentence down for a week.

Here is why it landed on me. I had always filed dignity under being respected. Being valued. Being admired.

But admiration is a verdict. Somebody looks at you, weighs you, and decides. Which means it can be withdrawn, and while you have it you are quietly performing to keep it.

Being needed works differently. My newborn does not appraise me. He has no opinion about my character. His claim on me is total and unearned, and that is exactly what makes it feel like dignity. I am not decorative in his life. I am load-bearing.

Which is why the opposite of dignity is not humiliation. It is being unnecessary. The retired man, the sidelined employee, the father who has been reduced to a wallet. Nobody insulted any of them. They were simply done without, and that is worse.

Then Zoey and I got to the harder question, which is what this means in a marriage, because two capable adults do not need each other the way a baby needs a parent.

The answer I keep landing on is uncomfortable.

The thing that kills dignity in a marriage is rarely disrespect. Most of the time it is competence.

Handling it alone. Deciding alone. Processing it alone. Never visibly needing her. It reads as strength and it lands as *I am fine without you*, which is the despair half of Brooks' sentence delivered by a man who thinks he is doing a good job.

I say this as the man who built an AI chief of staff so I would need fewer people.

And that is the connection I did not expect to make this week.

The loudest story in my industry right now is that humans are about to become unnecessary. Every week another chart, another projection, another confident post about which category of person is obsolete next. I think that story is wrong on the economics, and every previous technology wave says so too. New tools have consistently created more work than they destroyed, in categories nobody could name in advance.

But I think it is wrong at a deeper level than jobs.

Look at how God does it. He does not need anything. He is not short of capacity, and He never runs out of options. And He still hands out real assignments with real consequences, and lets people carry them, and waits. Christ asked a Samaritan woman for a drink He did not require. In the garden He asked three friends to stay awake with Him, and they failed, and He had asked anyway.

None of that was need. All of it was conferred.

Which means being needed is not only something you discover. It is something you can give.

Where have you been so competent that the people who love you have quietly stopped feeling necessary?

The News: Two positions I have written to you about are working, and both for the same reason.

ServiceNow is up 39.8% since I wrote about it in issue #020 back in June. Cloudflare is up about 21% over the same stretch and about 44% for the year, and its last quarter did $696 million in revenue, up 36%. I hold Cloudflare through 2027 LEAP call options. I put in around $60,000, took $30,000 off the table earlier this year when it ran, and then bought back in. The position is worth over $100,000 today.

The Backstory: In June I told you the uncrowded trade was not the stock the crowd was excited about, it was the good company the crowd was scared of. ServiceNow was trading at half its high at the time. That was the entry.

The Thesis: What I am watching is a repricing, and repricing is the part of investing I understand better this year than I did last year.

Earlier this year the market got frightened that AI would make software companies obsolete. People called it the SaaS apocalypse. It was not imaginary. Reuters reported on the fifth of February that the S&P software and services index was on track to shed a trillion dollars in about a week. By CNBC's numbers the main software ETF fell 24% in the first quarter.

The selling was indiscriminate, which is the opportunity. Some software companies genuinely are threatened by AI. But a few of the names that got dragged down are companies the agentic wave should help, because when AI agents start doing real work inside a company, they need somewhere to execute. Cloudflare unclogs the network. ServiceNow unclogs the work inside the building. Neither of those jobs gets smaller when agents show up. They get bigger.

The market has partly worked this out. That same ETF was back to roughly flat for the year by early August. But the recovery has been selective rather than universal, and there are still good companies sitting in the discount bin they were thrown into on a headline.

The Position: I am converting part of my Cloudflare LEAP calls into actual shares. The options did their job; the thesis is a decade-long one and I would rather own the company than a dated bet on it.

And I did open Coinbase. In the last letter I told you I was putting on a starter position, and I did, at about 1% of the portfolio. I intend to build that to somewhere between 5% and 10% over time.

Which brings me to the rule underneath all of this, and it is the only genuinely useful thing I have learned about investing in the last two years.

If I believe in a company long term, I buy a small piece immediately, even when I am certain it is not the right price.

Not because my timing is good. Because of what owning does to my attention.

I have lost more money to opportunities I identified correctly and then never acted on than to any position that went against me. I would spot the company, decide to wait for a better entry, and then stop tracking it entirely, because nothing was at stake. Six months later it had doubled and I had a note about it somewhere.

“Where your treasure is, there your heart will be also.” - Jesus

I have always read that as a warning. It also describes how attention actually works. Put money on it, and you will read the earnings call. Put nothing on it and you will read the headline.

So the small position is not really a bet. It is a subscription to paying attention.

Watching: Tesla. On Wednesday it put the Cybercab into service in Austin, with public rides opening Friday evening. No steering wheel. No pedals. No mirrors. Nobody in the front seat. For years the standard response to that car was that it could not be built or would not be allowed, and this week people got into one.

Tesla runs roughly 250 driverless taxis across six cities. Waymo runs about 4,000 across fourteen. Nevada regulators cleared Tesla in August to operate up to 5,000 autonomous vehicles in Clark County over the first year.

Manufacturing is the thing Tesla is genuinely world-class at, so if regulators keep moving, the ramp is the part I expect to surprise people. It just has not happened yet.

The Setup: I give Claude Code a problem. It goes away for thirty minutes. I am not going to watch a progress bar for half an hour, so I start something else. Then a second agent finishes and wants a decision. Then a third.

My attention is split all day. By everything I have ever read about focus, I should be performing terribly.

The Study: Researchers in Hong Kong trained mice to do two things at once and watched their brains while they learned, published in Neuron this month.

At first the two jobs fought over the same equipment. Same kitchen, two cooks, one chopping board. That is the interference everybody warns you about.

But with practice, the brains built each job its own station. The two tasks stopped sharing and started running side by side.

So the fighting was not a ceiling. It was the beginner stage.

The Honest Part: These are mice, ten of them, pushing a lever. Half never got the hang of it. And what I do is not really two things at once, it is one thing, then a wait, then another thing.

So this does not prove my day is fine. It just makes me suspect the rule I inherited was describing a stage, not a law.

As promised, my private notes. No fluff.

See you next week,

Gabriel Judah

P.S. The two-brain build in the first section, the personal one and the company one with a door between them, is what I am working through with members right now. Doors are closed at the moment while I take care of the people already inside. The waitlist is the only way in, and it gets told first when the next slots open. Join the waitlist here.