Process
Audit. Strategy.Build. Automate.
Step 01: Audit
A week inside your workflows and systems. What gets done, by whom, how often, and where the hours go.
- Shadow the people doing the work
- Map every system of record and its API
- Find the three or four steps worth removing
Deliverable
Written findings, ranked by hours saved.
Step 02: Strategy
A 90-day plan with the quick wins first, the build-or-buy calls made, and the guardrails agreed before anything runs.
- Quick wins first, platform work second
- What stays human, what gets logged
- Cost, hosting and data-residency decided up front
Deliverable
A plan any engineer could execute.
Step 03: Build
Production code from the first week. Small releases, real users early, tests and a deploy pipeline from day one.
- Weekly releases into a staging environment you can click
- Integrations against your real systems, with safe write-back
- Lint, type-check and tests on every pull request
Deliverable
A system in production, with the pipeline to keep it there.
Step 04: Automate
Agents take the repeatable work, with a human checkpoint where the decision matters. Everything logged, everything reversible.
- One-click approvals by email or form
- Overnight agent runs with a human merge gate
- Audit trail written into your system of record
Deliverable
Hours back per person per week, measured.
Engagements
How an engagement runs.
Three common shapes. Each one ends on a written deliverable, and none of them commits you to the next.
- Steps 01
A one-week audit
Step one on its own. A week inside your workflows and systems, then written findings ranked by hours saved. Keep them, act on them yourself, or bring me back.
- Steps 01 + 02
A 90-day plan
The audit, then the plan: quick wins first, build-or-buy calls made, guardrails agreed before anything runs. Written so any engineer could execute it.
- Steps 03 + 04
Build and automate
Production code from the first week, weekly releases into a staging environment you can click, then agents on the repeatable work. A human merge gate on every change.
Overnight
What runs overnight.
The Automate step, running every night across four codebases. A person merges in the morning.
A representative night for the overnight agents. Every line is a state the pipeline prints.
A ticket lands in a queue during the day, often by voice. Each night the queue is read, and a planning step checks out only the repositories those tickets actually touch.
An agent writes the spec with the ticket’s priority honoured, builds it in an isolated workspace, and runs lint, type-check and tests on every pull request it opens.
An automated review scores each pull request. It gets at most two fix commits, then it is handed to a person. Nothing merges without one.
Who you’ll work with
One engineer, start to finish. Twelve years on recruitment desks before writing code, and still shipping production code every week.
About JamesBook a call
Start with the audit.
You'll get a straight answer: a plan, a build, or “you don't need this.”