You already have the AI.
Right now it makes more work than it saves.
Not because you're prompting wrong. Because nobody gave the model the context, the rules, or the means to check itself. We spent a year solving that in the open, on a real codebase, and the fix turned out to be embarrassingly simple.
Most people are one well-structured document away from a completely different experience. We'll sit down with you and write yours.
Most of what you hate about AI is real. We built around it instead of pretending otherwise.
Sycophancy is the default failure. So the rules live in a file the model reads before it reads you — written once, calmly, by someone who was not mid-argument. It is much harder to flatter a document into changing its mind.
So nothing it claims counts until something re-derives it. In our codebase that is a bank of scripts — one for every rule we have broken twice. In a document set it is a self-audit pass. A claim is marked "asserted" until it is measured — and we write the difference down.
The only real fix is writing it down where the next session will look. Our documents evolve themselves: every session that learns something is required to leave the file better than it found it, or the work is not finished.
Volume is not the goal. Constraints are. Ours is a codebase where two people run an entire platform — and the thing keeping quality up is not the model, it is the short list of rules the model is not allowed to break.
We are not going to tell you AI is magic. It is a very fast collaborator with no memory, no judgment about your business, and a habit of telling you what you want to hear. Handled that way — honestly, with structure — it is the most useful thing we've ever worked with.
Where we stand on open models, privacy and artistsIt's a markdown file. That's the secret.
We're not going to pretend it's more than it is. There is no proprietary model, no plugin, no framework to install. There is a document, shaped a particular way, that turns a general-purpose model into something that knows your work. Almost nobody actually does this. The ones who do are having a very different year.
Bootstrap
One document that holds what is true about your work: the vocabulary, the constraints that must never be broken, the decisions already made so they stop being re-litigated. Shaped for how a model actually reads — the things that matter most at the very top and the very bottom, reference material in the middle where attention is cheapest.
Evolve
The file is not a snapshot, it's a ledger. Every session that discovers something adds it — a new failure gets a numbered entry, a new decision gets recorded so nobody argues it twice, a risk you're accepting gets written down with its cost. Numbers are permanent. Nothing is ever renumbered. Six months in, the document is smarter than any single conversation that built it.
Audit
A rule stated twice is a rule that will be broken a third time. So rules graduate into checks that run on their own and fail loudly. In code that's a script. In a document set it's a pass that re-derives every claim and flags the ones that stopped being true. This is the part that separates a nice prompt from infrastructure — and it's the part almost everyone skips.
Port
Here's the part that sounds too good to be true. A well-built bootstrap is not domain-locked. “Take this and port it to running my clinic” genuinely works, because what you built was never really about the subject — it was about how a model is told what matters. We've carried the same skeleton into documentation systems, research, personal life management, and a game engine.
The platform you're standing on is the proof.
This site, every product behind it — including the ones we don't advertise — the login you'd use, the database rules that keep one business's data away from another's. Two people, working this way. You can poke at all of it.
Counted from the repository in September 2026. They only go up, and we will not pretend otherwise when they do.
Right beside our customers' data, on the same servers, with no staging safety blanket underneath it. Which means a bad enough deploy is one we find out about by being unable to file the ticket about the bad deploy. It is a slightly absurd safety net. It is also the most honest stability test we know of, and we have been passing it for a while now.
Every product we make, one Next.js and Postgres monolith, each white-labelled onto the customer's own domain. A set of security rules that cannot be violated without a script objecting. Real businesses run on it today.
The same method pointed at prose instead of code: hundreds of pages that cross-reference each other, hold their own consistency, and get audited by the thing that helped write them. No code involved anywhere.
Built ground-up in a language and an engine, because the honest test of a method is whether it survives contact with something genuinely hard and unfamiliar. It did.
You don't have to be technical. You have to have real work.
Small business owners
Intake forms, client notes, quoting, the same five emails you rewrite every week. The unglamorous stuff that eats your evenings.
Teams shipping software
You have AI writing code and no structure holding it accountable. This is the part you are missing, and it is the difference between speed and cleanup.
Writers and researchers
Long documents that must stay internally consistent. Research that has to be traceable back to a source rather than vibes.
People who just want their life organised
Genuinely. The same structure that runs our platform runs a household. It is not a metaphor, it is the same file shape.
If you want someone to tell you AI will replace your staff, or to sell you a strategy deck about it, we're the wrong call. We work on the thing in front of you, on your screen, and you leave with a file you own and understand. That's the whole product.
We mean that more strongly than this box doesPriced by the outcome, not the hour.
Most of this work is short. That's the point of it. So the question we care about isn't how long we sat together — it's whether you're still using what we built a month later.
Working Session
We get on a call, screen shared, and work on the thing that's actually bothering you. No slides, no discovery phase.
- ✓Your real work, live, not a demo
- ✓You keep every file we write
- ✓A recording and a written summary
- ✓Follow-up questions by email, free
The Bootstrap
We learn your domain, write your bootstrap file, and sit with you while you use it. Then we revise it after a week of real use.
- ✓A working session to learn your domain
- ✓Your bootstrap, written by us, yours to keep
- ✓A second session once you have used it
- ✓One revision pass after real-world contact
- ✓Self-audit structure built in from day one
Ongoing Advisory
For teams. We stay in the loop, review what your structure has grown into, and keep it from rotting.
- ✓Eight hours a month, used however you like
- ✓Async review of your files and prompts
- ✓Reply within a day, from a human
- ✓Cancel any month, no term
We mean that literally and we've never had to honour it. Team workshops and custom engagements are quoted separately — tell us what you're trying to do and we'll tell you what it takes.
Tell us what you're stuck on.
One email, a couple of sentences about your work. We'll tell you honestly whether an hour with us would change anything — and if it wouldn't, we'll say so.
No sales calls. No demo requests. Just humans talking to humans.