Your team ships more code with AI than a year ago. A surprising share of it comes back within three weeks to be fixed, reverted or rewritten. I find out why, change the workflow, and prove it on your own numbers.
Nine in ten developers now use AI tools daily, and trust in what those tools produce has fallen to one in three. Nearly half say AI-written code takes longer to debug. The dashboards you may already pay for show the symptom as a rework rate. They do not tell you what to change on Monday. That is the job I do.
Who this is for
- You lead engineering at a company with roughly 10 to 200 engineers.
- Your team uses Claude Code, Cursor, Copilot or similar tools, and you suspect the gains are smaller than the invoices.
- Review queues are longer, reverts are up, or seniors are quietly rewriting junior pull requests.
- You want a number that moves, not a slide deck about transformation.
What you get
Step one: the audit. Two weeks, fixed price, no hourly meter. I read your pull requests from the last 90 days, compute the rework rate on AI-assisted changes from your git history, interview four to six engineers, and review how your tools are configured, including permissions and secret handling. You receive a written report with a baseline, the five things causing the most rework, and a 30-day plan. Read a sample audit before you decide.
Step two: the pilot. Thirty days on one team. I set up the review gates, prompt and skill libraries, test generation, and evaluation harnesses the audit calls for, and train the team to run them. We agree on one metric before we start. If the rework rate on AI-assisted changes has not fallen by at least a third at day 30, the pilot is free.
Step three, if you want it: the retainer. I roll the pilot out to the rest of the organisation, keep the metrics honest as the tools change, and leave your team able to run it without me. The point of the name is that I work inside your team and then disappear.
How it is measured
Rework rate is the share of changed lines that are modified again within three weeks. It is the measure the DORA research programme uses, it is computed from commit history alone, and if you already run GitClear, LinearB, Swarmia or DX it is the number on your dashboard. Their tool is the referee. I never grade my own work.
I take the baseline from your repository before I change anything, so the before figure is yours, not mine. During the pilot the number is reported weekly. If it is not moving by week two, you find out in week two.
Pricing
- Audit: $6,000, fixed. Two weeks. Credited in full against a pilot.
- Pilot: $12,000 for 30 days on one team. Free if the agreed metric does not move by a third.
- Retainer: from $8,000 a month, three-month minimum, cancel at any month end after that.
The first three audits of the quarter are offered at $4,000 in exchange for a named, numbered case study.
Why me
I wrote Debugging Is the Skill, an 88-page method for finding faults in code you wrote, code you inherited, and code an AI wrote for you. This service is that method pointed at a team instead of a program. I write about the same problems every week on The Haunt, so you can read how I think before you pay for it.
What I will not do
- Sell you an open-ended hourly engagement.
- Install an autonomous agent and call it a transformation.
- Promise your team will ship faster. I do not control your roadmap. I do control rework.
- Touch production credentials, or keep any of your code after the engagement ends.