How we build · Field notes from the build of Genacy
Genacy was built by one person using AI models as the engineering team. This is the first-person account of how that actually went: the month with no code, the day the documentation started lying, the bug that hid behind green tests, and the afternoon a tool reported success while writing nothing at all.
The founder sets the direction. AI does the engineering. Checks feed the next session.
Agree the workflows and constraints before writing code.
Work in bounded sessions with clear handovers.
Review structure, behaviour, content and the real browser.
Carry failures and lessons into the next session.
Read them in order, or start with whichever sounds most like your week.
Thirty-five workflows and more than a thousand decisions, locked before anything was built. It sounds like procrastination. It was the best decision of the project.
Read Act 1 →Notes that drifted from the code, a rename that took eleven decisions for one word, and a bug no test could see.
Read Act 2 →Four edits, four success messages, zero bytes written — and the boring habit that got the work back.
Read Act 3 →Everything else was negotiable. These were not, and they are why one person can move this fast without breaking things quietly.
It knows only what you put in front of it. The documents describing the project drift away from the project unless you force them back together, every session.
Does it parse. Does it behave. Is anything lost. And what does a real browser actually show. None of them replaces the others — we learned that one twice.
Verify the result yourself: the bytes, the render, the count. One tool reported success four times while writing nothing at all.
Every lesson becomes a rule the next session has to follow. A method you keep in your head is a method you lose when you are tired.
This is the advice nobody wants to hear and everybody needs. Before the interesting work starts, build the scaffolding: the house rules the AI reads at the start of every session, the skills it can call on, the trackers that say what is done, a lessons-learnt document that grows, test cases generated rather than hand-written, and permissions set so the tools can do their job without being able to do damage.
It feels like a day lost. It is the difference between an AI that compounds your work and one that quietly undoes it while sounding pleased with itself.
A file the AI reads every session: how we work, what is banned, what to do when unsure.
One place that answers "where are we up to", generated from the plan rather than from memory.
Every mistake written up as a rule, carried forward into the next session.
Test cases generated automatically; tool permissions scoped so nothing can be quietly destroyed.
Coming soon · First 2,000 trial users only
Help shape Genacy and keep the early-member benefit for life.
Join the beta trial