The Private Apprentice

Last week Block shipped the fastest-adopted tool in its history, a way for anyone to build their own software, where the platform owns what makes it safe. It’s the same bet I’ve spent the last year building Foundry on. Here’s the bet, and the two places I went a different way.

The person AI keeps failing

There’s a particular kind of person that the last three years of AI has mostly failed: the expert whose work is too important to get wrong and too private to paste into a chatbot. The financial advisor with twenty client reviews to prep. The fractional CFO closing five books at once. The estimator assembling a bid from takeoffs, historical costs, and twenty years of judgment that lives only in their head.

These people don’t need a smarter chatbot. They’re not under-tooled, they’re overworked. They need something closer to an apprentice: someone who watches how they work, learns the parts that repeat, and quietly starts taking the busywork off their plate, while the expert stays the one in charge.

That’s what I’ve spent the last year building. It’s called Foundry, and the easiest way to explain it is to start with what Block just did.

What Block just proved

Block, Jack Dorsey’s company, rolled out something internally called App Kit. The idea: a non-engineer describes a tool in plain English, an agent builds a working app, and it’s running within the hour. The clever part isn’t “AI writes code.” It’s the split: the agent does the building, and the platform owns everything that makes the result safe and durable. Jack called it the fastest-adopted tool in the company’s history.

That split is the whole game, and it’s the thing I think most people are still missing about this moment. The model stopped being the hard part a while ago. The hard part, the part that decides whether normal people can actually trust this with real work, is the safe, durable, accountable layer around the model. Block proved the appetite for that is enormous. They proved it inside one company.

What Foundry is

Foundry is a private apprentice. You work the way you already do. It quietly builds a private memory of your practice, your clients, your cases, your decisions. It learns the things you do over and over. And it turns those into playbooks: named, repeatable jobs it runs for you, with your approval. Over time it gets measurably better, and your work never leaves your control.

None of that requires you to learn a new tool, wire anything together, or trust a vendor with your most sensitive data. The platform handles the unglamorous, load-bearing parts, remembering, running, measuring, keeping it private. You bring the expertise. That’s the same split Block found, pointed at a different kind of person.

The two places I went the other way

Block built App Kit for Block, for its own employees, for building internal software, on its own managed cloud. That’s the right call for them. I made two deliberately different choices, because I’m building for a different person.

First: it’s for everyone, not one company’s staff. The people who most need an apprentice, independent experts, regulated practitioners, the person doing skilled work alone, are exactly the people who get served last, because no one’s going to build them an internal platform. Foundry is the thing you can just have.

Second: it’s built for people who can’t afford to be wrong. So the platform owns more than “safe and durable.” It owns the memory of how you work, the measurement that proves the apprentice is actually helping, and the privacy. Block’s layer is managed cloud; Foundry runs private, with a fully-local option, because the people I’m building for handle things they can’t hand to anyone, and that’s a deliberate design choice, not a knock on managed cloud.

Block proved the demand inside one company. The harder, more valuable version is making that same idea available to everyone, private, measured, and useful well beyond writing code.

Who it’s for

I think about who this is for as types, not industries. There’s the independent expert, anyone who repeatedly assembles a complex, high-stakes output from scattered, private inputs (a proposal, an intake, a close, a report) and can’t paste it into ChatGPT. There’s the individual who just wants to remember everything and carry nothing. And there’s the organization that needs this built into how it actually works, with the audit trail a regulator can read.

Whether you’re an advisor, an estimator, a fractional CFO, or a clinic, if your work has that shape, the apprentice fits.

The honest part

I’d rather tell you exactly where this is than oversell it. What’s real today: the apprentice captures your work, builds a private memory of it, and makes it genuinely easy to turn a repeated job into a playbook it runs for you. The part where it notices the pattern on its own and proposes the playbook before you ask, that’s the most exciting piece and the one I’m still building. I’ll show it when it’s real, not before.

That honesty is part of the point. The whole reason an apprentice earns its place is that you can check its work. You can’t control what you can’t measure, so measurement isn’t a feature bolted on; it’s how the thing earns your trust.

Get your time back

That’s the whole idea. The expert work stays yours. The toil around it gets done. And your world never leaves your control, it’s private, never trained on your data, with a fully-local option, and you can delete or export everything anytime.

Block built an apprentice for one company. I’m making one anyone can have. If that’s the kind of thing you’ve been waiting for, get on the list, I’d genuinely love to hear how your work is shaped.