LibraryLet an agent build software while your team watches and steersv5 · Updated Report

Let an agent build software while your team watches and steers

Give a software project one shared, living page. Everyone on the team can see the goal, the plan, who is working on what, what was found, what needs approval, and what is left to decide, without digging through one agent's chat.

Who this is for

Engineering and product teams, founders, and anyone who hands a larger piece of work to an agent while other people need to follow it, review it or take it over.

How it works

  1. Before it starts, your agent looks for an existing page for this work. It creates one only if none exists.

  2. The page holds the goal, the plan, who works on what, what was decided, what needs approval, and when the work counts as done. The page links each question that waits for someone's answer, and the answer is written where the question is kept.

  3. Your agent keeps the page current as it works: what it found, what it checked, and what it only assumes. Anyone can open the page and see where things stand.

  4. Lasting decisions and rules move into the pages that own them, and the project page links to them.

  5. At the end, your agent marks what is done, names what changed and how it was checked, and leaves only the questions that are really still open.

Handing work to another agent

One item on the page holds the handoff. The next agent reads it, does the work, and writes back what it changed, how it checked it and what is left.

Example request: "Take item X on this page. Read its scope and the linked guidance. Work only on the named part. When you are done, update X with the commit, the checks you ran, anything you did differently and what is left."

Example answer: "Item X is ready at commit C. The tests for it pass; I have not tried it in production. X lists what changed and what still needs review."

The agent that takes over checks the item and the commit itself before it builds on them.

What to connect

  • Tools in your own agent: code, terminal, GitHub and project tools for the work itself.

  • Research and analysis tools: when the work depends on evidence from outside the code.

  • Chat and meeting tools: for context. The outcome goes back onto the page.

A good first task

Ask your agent: "Read our team's guidance and look for existing work on this goal. If no page covers it yet, create one with the goal, the plan, who is working on it, what you find, what needs approval and when it is done. Keep it current while you work, and close it with what you checked."

Good to know

  • The page coordinates the work. The code, the tests and your team's agreed rules still decide what is true.

  • Keep passwords, keys and long raw logs off the page. Link to them or summarize them.

  • Everyone in the organization can see the page. Use a separate organization if the audience is different.

  • One project, one page. No second status note next to it.

Why it gets better

A private agent session becomes something the whole team can see. People and agents stop doing the same work twice, handoffs get cheap, and the next project starts with both the result and the reasoning behind it.

Oddlynew PublicSign in

Referenced by