b.akery

Plain-language contracts, a correction ledger, and release gates you can run: how a non-engineer owns software that AI wrote.

STARTER KIT SHIPPED the seven working components below, gates proven before they ask for trust

LIVE DEMOS SHIPPED a playable game, a voice agent that calls you back, and the gates running live on the evidence page

AI writes the code. You still own the release. What you need is evidence you can judge without reading every line — and a system built to produce it. The kit is that system: rules the build must honor, a ledger that turns every failure into a permanent rule, numbers that carry receipts, and gates that block unverified work from shipping.

The kit runs on itself. Its correction ledger records the real failures of its own construction, and those rules now run as automatic checks on the kit itself. The standard everything serves: a passing release command proves the promises encoded in the contract.

GET THE FREE KIT (ZIP) READ IT AS ONE FILE BROWSE THE REPO

The seven working components

Standing contract — the facts your software must never get wrong, written once, enforced every session
Correction ledger — every escaped failure becomes a dated, permanent rule
Calculation provenance — no number ships without the record that reproduces it
Release gates — checks that fail the build, each proven against planted errors
Render-and-look log — dated proof a human looked at what shipped
Recipe cards — recurring chores your assistant runs by name
Credentials guide — holding your own keys without pasting a secret into a chat

Best fit: websites, internal tools, workflow automation, and prototypes. The gates improve consistency and stop caught mistakes from returning; they do not replace security review or professional oversight for systems that move money, hold sensitive data, or can hurt someone when they fail. Narrow claims are the point.

Fastest start

Unzip the kit, open INSTALL.md, and paste the prompt into the AI session you already use for your project (one that can read and write files: a coding agent or desktop assistant, not a plain chat window). Your agent does every technical step; you answer business questions in plain language, and the gate proves it works by failing once in front of you.

Need it adapted to an existing project, or operated for you? Work with the author.

The honest part

This method has one longitudinal case study: its author. The kit's retirement rule, written before any second case existed, defines exactly what would let it claim more, and puts that evidence under the subject's control, not the author's. It ships inside the kit (read it here, under instruments).