Engineering productivityworkflow
Async dev via Slack-mentioned agent (Codex/Claude in a dedicated channel)
Problem solved: Feedback piles up faster during client calls and stand-ups than anyone can implement; mentioning an agent in Slack converts that backlog into ready-to-merge PRs by the end of the meeting.
What
Install the Codex Slack app or Claude Code's Slack integration. Whenever feedback or a small task surfaces in a meeting, drop it into a dedicated channel and @-mention the agent. It picks up the message, spins up a branch per task in the background, and files a PR. Slack becomes the queue; the laptop stays closed.
Why it matters
Reported one-shot rate around 60% on small tasks. The asynchronous queue means a 60-minute call can produce 10 PRs in parallel without anyone touching a terminal. Larger tasks land as draft branches a human can finish.
End-to-end
- Create one dedicated Slack channel per project for agent dispatch (
#agent-projectX). - Install the Codex Slack app or Claude Code's Slack integration; authorize against the repo.
- During calls, type or paste each feedback item as a fresh message and @-mention the agent. Keep one task per message.
- Let the agent fan out: one branch + one PR per mention. Don't expect a back-and-forth — these dispatchers are one-shot.
- After the call, triage PRs: merge the clean ones, leave comments on the half-baked ones, claim the rest for a human pass.
- Use Codex for async/long tasks; Codex's strength is that it works in the background. Use Claude Slack when the task needs more reasoning per shot.
Gotchas
- Codex Slack is one-shot only. Don't expect a conversational loop in the channel — write each task as a complete, standalone instruction or you'll get a PR that solves the wrong problem.
Tools
- Slack workspace, repo with PR-based workflow
- Codex CLI or Claude Code with the official Slack app installed
- One Slack channel per project + clear naming convention so @-mentions don't cross-fire
← All casesB-119