Two weeks. Your agents' prompts, their output and your tests. Back comes the constitution and the invariants your system is missing.
For teams putting AI-written code into production. I read what your coding agents were told, what they produced, and what your tests actually prove, and I hand back two things most AI-built systems lack: a written constitution, the rules the agent must obey with the reason behind each one, and pinned invariants, tests that prove the forbidden thing cannot exist anywhere in the code.
$15,000, fixed. Credited in full against engineering work booked within 90 days. Book a 30-minute scoping call.
What you get
The constitution
- The rules your agent must obey no matter what it is asked, each with its reason and the failure it prevents
- A standing-landmines section: the settings one change away from a silent breach
- Written to be loaded into every session, in the file your agent already reads
Pinned invariants
- The list of things that must never exist in the code: a second order path, a hardcoded exchange parameter, a default that points at production
- Tests that parse the source and fail if any of them appears anywhere
- Run in your CI, so a future session cannot reintroduce them
The findings
- A severity-scored report on the agent's output and your tests, with a fix order and effort per finding
- A one-page summary for the CTO
- A live readout of 90 minutes, and a re-check of the critical findings once you have fixed them, within 30 days
How the two weeks go
A free 30-minute scoping call. Scope, access and dates in writing. Read-only access under NDA; I keep no copy of your code after the engagement.
I read the prompts and rules files, the last few weeks of agent output, the tests, the CI gates and the deploy path. I ask questions in writing and stay out of your way.
I write the constitution, the invariants and the pinning tests, and run the tests against your code so they fail where they should.
The readout with your engineering lead, then revisions. You own every document and every test from that day.
The price, and what it does not cover
$15,000, fixed, for one codebase and a team of up to about twenty engineers. Half on signing, half at the readout. Work starts within two weeks of signing. If you book engineering work with me within 90 days, the full fee is credited against it.
Not included, so nobody is surprised: a penetration test, a performance audit, a rewrite of your code, or a replacement for your own code review. Larger codebases or regulated scope, including trading systems, are quoted after the scoping call, usually up to $25,000.
No outcome is guaranteed, because no honest reviewer can guarantee one. What is guaranteed: the price, the start date, your ownership of the report and the tests, and the re-check of critical findings.
A good fit
Teams shipping code written by Claude Code, Cursor, Codex, Copilot or similar agents into production, where a defect costs money, data or trust. One to twenty engineers. A lead who wants rules that hold when nobody is looking.
Not a good fit
Prototypes that will be thrown away. Teams that want the agent to run without review. Anyone looking for a certificate rather than a set of mechanisms.
Where this comes from
In 2026 I guided an AI coding agent to build and run Argus, a platform for fully automated trading strategies, under exactly this discipline: a 741-line constitution read at the start of every session, 87 decision records, 3,540 tests including the ones that parse the source, and a kill switch I pressed from my phone. It traded a real brokerage account. Its first strategy was shut down on pre-registered evidence. The method, the defects the loop caught and the ten rules I would give a CTO are in part one, and a reference implementation is public on GitHub.
Before that: 25 years of trading infrastructure, including the order-entry and market-maker gateways at the New York Stock Exchange and the market-connectivity, FPGA and DevOps teams at Chicago Trading Company.
Start with a scoping call
Thirty minutes. Tell me what your agents are building and what a bad day would cost. You leave with a written scope and a date, or with a straight answer that this is not the right review for you.
Nothing on this site is investment advice. The trading strategy in the write-up lost money and was shut down.