You bought the tools. Let’s build the practice.
Your team has Claude Code, Cursor, Copilot, probably more. The seats are paid for either way. What is missing is an agreed way of working with them — so your best engineer is reverse-engineering it alone at eleven at night, and what they learn on Tuesday reaches nobody.
The audit
Priced in the conversation, not on the page.
Two days on site · one-time · fixed fee, agreed before we start
Two days with your team, on your repo. I score how you work with agents today against six dimensions, and hand back the baseline plus a ranked list of changes with the effort and the expected payoff against each. It is yours whether or not we carry on.
If the readout does not turn up at least three changes you agree are worth making, you do not pay. The invoice follows the readout, not the conversation that set it up.
- 01Context engineeringIs there a CLAUDE.md or AGENTS.md, is it current, and does it describe the repo as it is rather than as someone hoped it would be?
- 02Tool fitWhich tool for which job, written down. Or four subscriptions and personal preference.
- 03VerificationWhat proves an agent’s output is right before it merges. Tests, types, build gates — or a tired human reading a large diff.
- 04Task decompositionHow work reaches an agent. One-shot prompts, or specs, plans and checkpoints.
- 05Shared practiceWhether what one developer worked out on Tuesday reaches the other six, or everyone rediscovers it alone.
- 06Feedback loop speedWall-clock from change to signal. The one dimension that is already a number, so it gets measured rather than scored.
Each is scored zero to five against written, observable evidence, so two people looking at the same repo land on the same number. The sixth is measured rather than judged, because it is already a number and a consultant’s opinion with numbers painted on it is not a baseline.
The retainer
A monthly fee, agreed after the audit.
Two days a month · cancel any month
- One day with the team. Pairing, setup, review, and making the changes from the audit actually stick. Remote, except every third month.
- One day remote. A read on what changed that is worth your attention, written for your stack rather than for everyone, plus answers to whatever came up.
- Every quarter, on site: we re-score. Same six dimensions, same rubric. That is the number that says whether this is working, and it is the reason to stop paying me if it is not.
Larger engagements, work delivered rather than advised, and public-sector procurement all exist too. All of it is scoped and priced the same way — against your team and your repo, in a conversation, rather than off a page that has met neither.
The same method runs on our own work, in public. Every experiment gets written up with the number attached, including the ones that fail. The first records are landing now.
Read the log before you talk to me. It is the honest version of what I would otherwise have to claim.
Read the log firstContact us
Tell me what your team builds and we will work out whether this is a fit, and what it costs. If it is not a fit, I will say so rather than sell you two days. Or grab twenty minutes and we will do it out loud.