One operating manual for every agent you work with.
Point them at the same instructions and the same operating loop: sense, structure, act, know. Agents write to the record as they work — progress, evidence, blockers — and humans keep every decision.
This website's newest pages were built by an agent running this loop.
Every agent in our studio — IDE copilots, terminal agents, background workers — onboards through one manual. They pick their task from Staff, cite its reference code in every commit, verify their own work, and file the evidence back. There is no separate “AI workflow” here; agents are coworkers with credentials, and the entire studio state — including this site’s content pipeline — lives in the same system they write to.
Sense, structure, act, know.
Four steps that turn a language model into a colleague: it starts from the real state of the organization, and it leaves the state better than it found it.
Agents observe, not guess
A session begins by reading the organization: active projects, open tensions, the task board, governing knowledge. Context is fetched from Staff — never improvised from a chat history.
Work gets an address
Before code moves, the work has a project, a task, and a reference code. Ambitions become trackable objects with owners, priorities, and acceptance criteria.
Implementation with verification
Agents build, then prove: builds pass, tests run, pages render in a browser. “Looks right” is not a state — the definition of done is an observed passing check.
Results become records
Progress, evidence, blockers, and the lessons worth keeping are written back to Staff. The next agent inherits a current state of the world, not a mystery.
Uniform behavior, without uniform thinking.
Same rules, same knowledge, same record — while every agent keeps the strengths you chose it for.
The unglamorous difference.
- Prompts copy-pasted between chat windows, improving nothing
- Context that dies at the end of every session
- Agent work you can’t trace back to a business objective
- Onboarding each new agent by explaining the company again
Your agents stop being clever strangers. They become staff: consistent, briefed, accountable, and on the record.
Your next agent shouldn't need onboarding.
It should need credentials. Give every agent the same instructions, knowledge, and rules your team works by.