For engineersGuide
Initiatives from your agent
Say you’ve worked an idea through with Claude, ChatGPT or Claude Code over a few days, and kept it in topics as you went. It’s ready to be built. Ask your agent to start an initiative from those topics, and Emargen takes it from there. It writes the brief and the requirements, and sets how much each matters. Then it chooses an approach and cuts the slices. It stops at a Kickoff that’s ready to publish. Publishing it, and approving it, stay yours, in the app.
Your agent can only do what you could do yourself.
Before you start
Section titled “Before you start”- Your agent is connected, and may start and message initiatives. Those two tools come only with Everything, or Custom with them ticked. You also get reading with Read only and Topics and notes. See The tools.
- What you decided is in your topics. Emargen’s agent reads only what’s in Emargen, not your conversation with your assistant. Before starting, your agent adds anything you agreed that isn’t in the topics yet.
Starting one
Section titled “Starting one”Ask in your own words: “start an initiative from the region topics, and let Emargen shape it”. In Claude Code, the connection’s initiative-from-topics prompt asks for the same thing. Your agent finds the topics, checks with you which ones it draws on, and writes in what’s missing. Then it starts the initiative as you, the same as New initiative in the app. It gives a title, the topics by key, and a first message saying what to build and why. It’s yours alone unless you ask for everyone to read it.
Emargen first looks for an open initiative that already aims at the same thing. If it finds one, nothing is started, and your agent tells you which. It starts yours only when you say how it’s different.
You get back the initiative’s key and address. Your agent also names any topic only you can read. The brief can use those topics, but your colleagues’ chats can’t read them, so you may want to share them.
Emargen shapes it on its own
Section titled “Emargen shapes it on its own”From the first message, Emargen’s own agent shapes the initiative in its chat, a step at a time, the same way it would with you in the app. It reads the topics in full, what your organisation connected, and your code. Where the record doesn’t answer a question, it decides, and starts its reason with “Assumed:”, so you can find every call it made for you. It doesn’t run spikes. What nobody knows stays an open unknown, for you to see before you publish.
Nothing in the app says an assistant started it. We left that out on purpose: nobody wants their initiative to read “Claude for” their name. You own the initiative, and Emargen’s agent is named on the changes it makes.
The run stops when the Kickoff is ready to publish, or when only you can take the next step. It also stops if something fails, if it reaches a turn or time limit, or if two turns in a row change nothing. If you write in the initiative’s chat in the app, you’ve taken over, and it stops too.
Following and steering it
Section titled “Following and steering it”Ask your agent where it stands, and it reads the initiative for you. It tells you the step it’s at and whether the run is going, and sums up what’s written so far. Then it tells you what rests on assumptions, and what’s yours to decide before you publish.
To add something, say it. Your agent passes your words on. While the run is going, Emargen’s agent takes them into its next turn. Otherwise Emargen’s agent answers them in your chat in the app, and your agent reads you the reply. Say “stop” and the run stops where it is. Say “carry on” and it starts again, with a fresh count of turns.
When it stops ready, open the initiative’s Kickoff in the app. Read what rests on assumptions, change what you’d decide differently, and publish it when you’re happy.