Welcome to wit Discussions! #11
Replies: 1 comment
|
I usually run one coordinator plus two to four bounded coding-agent workers, mixing Claude, Codex, and Gemini depending on the task. The recurring conflicts are broader than simultaneous file writes:
The controls that have helped most are an authoritative shared task graph, explicit ownership of exact files and git operations, re-reading a file immediately before editing it, and keeping final integration with the coordinator. Worktrees isolate branches, but they do not solve duplicated work or stale assumptions. For small changes, exact file locks in one working tree can be simpler; for independent implementation streams, worktrees plus a deliberate integration gate are safer. Maintainer disclosure: I’m building Better Agent, a related local web workspace for supervising multiple Claude, Codex, Gemini, and other coding-agent sessions: https://github.com/ofekron/better-agent It is source-available and free for non-commercial use; commercial use requires separate permission. I have not used Wit yet, but its coordination protocol is aimed at a real gap. The protocol will be most useful if it represents not just file ownership, but task status, dependency changes, and an explicit “verified” handoff distinct from “agent exited.” |
Uh oh!
There was an error while loading. Please reload this page.
Welcome to Wit Discussions
This is the place to talk about multi-agent coordination, ask questions about Wit, and share how you're using it.
Some things worth discussing:
If you're here, introduce yourself:
What AI coding tools are you using? How many agents do you run in parallel? What kind of conflicts do you deal with?
We're early (v0.1.3) and actively building. Your feedback directly shapes what gets built next.
All reactions