How agents work together: SysMLEdge and Atelier

When you hand a job to a team of AI agents, two things have to stay straight: what's true about your system, and who's doing what right now. SysMLEdge holds the first and Atelier holds the second.

SysMLEdge keeps the truth. Your system model lives on a desk. Agents read only the part they need and suggest changes as proposals. Nothing changes until a reviewed proposal is saved, one at a time, so the model never splits into competing versions.

Atelier keeps the work. Each job gets its own session, a shared notebook the agents read and write. It holds four kinds of entries: your ask, the tasks, blockers, and done notes. Agents claim a task before starting, so if two agents claim it at once, one claim is refused. They read only what's new since their last look, so no one reloads everything.

The job runs as a loop. Your request is written down, and the agents confirm it with you if it's unclear. They split the work into tasks and do it in parallel. Then they check the result against what you asked for, and go around again until it's done. If the same thing fails twice in a row, they stop and ask you. You decide when the job is finished.

Why it matters. Agents stay in step without re-reading everything, your model stays the single source of truth, and you can always see what was asked, who's on it, and what's left.