Coding agents

What the codedoesn't say.

Claude Code and Codex decide things on their own while they work. They leave things for later and run into gotchas. None of it shows in the code. Emargen keeps it: as a note for the engineer who ran the session, and as a question or a change request for the people who planned the work.

Who sees what.

One afternoon at ACME. Jonas builds the second slice of Lena's plan, "Approve within your warehouse", with Claude Code.

A taskClaude Code claims "Check the approver's warehouse", builds it and closes it with merge request !482.Lena and the team see the slice move.
A note"I kept approvals in the existing audit table and added a warehouse to each row. Rows from before today have none."Only Jonas sees it.
A question"Should warehouse managers see requests from before they got the role?" The plan doesn't say, and Jonas can't decide it.It goes to Lena's inbox.
A change requestA warehouse on each audit row changes the weekly export two clients import. Emargen writes two options.Lena decides.

For engineers

What your agent decided without you.

When Claude Code or Codex works with you in a repository, it leaves you a note for what you should know and the code doesn't say. One note per subject, written as it happens.

Decision
What it chose on its own, and why.
Todo
What it left for later.
Gotcha
What would cost the next person an afternoon.
Idea
What's worth keeping, but not now.
Tomorrow starts where today stopped.When your agent starts in a repository, it's handed your open notes there, so it knows what it left and why.
Nothing said twice.Before it writes, the agent searches your notes. When it finishes one of your todos, it closes it.
Look back over last week.Ask your agent to look back, and it reads your earlier sessions for what it should have noted. The sessions stay on your machine.
Only you see them.Not your team, not your manager, not the administrators.

A note can become a plan.

A note stays yours until you turn it into something others can use: a page to work it through, a new initiative with its brief drafted from your notes, or a change request on the plan you're building.

Yesterday's notes on one page, every morning.

Ask an agent once, in your own words. Every weekday it reads your notes from the day before and writes you one page: what was decided, what's left, what to look at first.

For product managers

When the code can't follow the plan, you decide.

Coding agents take their tasks from the slices you approved. They add the tasks they find and close each one with its merge request, so you see the work move without asking anyone where it stands.

When only the plan's people can answer something, the agent asks them, in their inbox. When what's being built can't meet what was planned, it proposes a change. Emargen writes the options; the plan's owner chooses. The plan changes only when a person decides.

Keep the agent you already use.

Claude CodeCodex

Each person connects their own agent once, and it works in Emargen as them: it reads only what they may read, and what it writes carries their name.

Emargen doesn't write your code or run your agent. It gives your agent the plan, and keeps what your agent tells you.

The clients and the people here are invented to show how the work happens.

Agents dothe legwork.People take decisions.

Ask for access