How we work
A coding agent is a model in a loop with tools. Everything else, what it reads, what it may touch, what checks its work, where it runs, is the harness. That is what we build.
Birgitta Böckeler splits a harness into guides and sensors: guides steer the agent before it acts; sensors check its work after. We add the two things a company needs before it lets agents near production: an access model that keeps credentials out of reach, and infrastructure that stays reliable when change volume goes up a hundredfold.
01 Guides
An agent starts every session cold. What it knows about your system is whatever the repository tells it, in the order it finds it. We arrange the code so the right context is close to the work: clear module boundaries, one obvious place for each kind of thing, and short operating manuals at the root and in each area that needs one.
Then we write the guides: rules for the conventions your team enforces in review, skills for the tasks agents repeat, and specs that say what to build before anyone builds it. Guides are cheap to write and easy to overdo, so we keep them short and delete the ones that stop earning their place.
What you get
02 Sensors
When agents write the code, the bottleneck moves to knowing whether it is right. A team that reviewed twenty pull requests a week cannot read five hundred. The answer is not to read faster; it is to build checks that run on every change and say plainly what broke.
We build those sensors: fast unit and integration tests, end-to-end QA sweeps against a real stack, holdout scenarios the agents never see, review agents with narrow briefs, and merge gates that refuse what fails. This is the part most teams skip, and it is the part that decides whether the factory ships or just types.
What you get
03 Capabilities, not credentials
Anything in an agent’s context window can end up in a log, a transcript, a commit or a prompt to someone else’s model. So secrets never go there. Agents call tools that hold the credentials; the agent gets the result, not the key.
We set up a vault that brokers secrets to processes rather than people or prompts, MCP servers scoped to what each agent needs, separate keys per host and per project, and an audit log that says who used what. Agents get everything they need and nothing they should not have.
What you get
04 Orchestration
One agent in one terminal is a tool. Dozens working the same repository at once is a team, and it needs what a team needs: separate workspaces, a way to hand off, rules about who merges what, and somewhere to run that is not your laptop.
We set up agent hosts, isolated worktrees with their own running stack, shared MCP servers and skills, and the triggers that start work without a person in the loop: an issue filed, a check failed, a nightly sweep. The modular architecture underneath is what keeps parallel work from colliding.
What you get
05 Infrastructure
More change sets means more chances to break production. The bar does not move: customers expect the same reliability they had when you deployed on Thursdays. What changes is how you meet it.
We build infrastructure agents can use safely at scale: reproducible environments, deploys that prove which build is serving, progressive rollout, one-step rollback, and observability an agent can query when it is asked why something broke.
What you get
06 Engagements
1–2 weeks · fixed fee
We read your repositories, pipelines and access model, and tell you where agents will stall, where they will break things, and where they could already run.
Fixed scope
We build the parts the audit found missing: the harness, the verification, the access model and the infrastructure, against a scope agreed up front.
Ongoing
Principal-engineer or CTO-level time, every week, for teams that want someone senior steering the factory while it grows.
07 The usual sequence
An hour on a call about what you ship, who ships it, and where it hurts. No deck.
One to two weeks inside your repositories and pipelines. You get written findings and the first commits.
The missing parts, built against a scope we agree on before work starts, delivered as pull requests.
Your team runs the factory without us. If you want someone senior to keep steering it, we stay on, fractional.
→Lights on
Tell us what you ship, how many people and agents work on it, and what keeps breaking. Eric reads and answers every one himself.