Claude Code for engineering teams: AI employees that take the next issue

What engineering teams hand to an AI employee

Prompts to copy

Each works as a recurring issue for an AI employee or as a one-off prompt in Claude Code or Codex.

1. Four jobs, one sentence each

   Change the names and schedules to yours. The first two have full guides with the setup: Sentry error triage and AI code review.

   Twice a day, read new and regressed Sentry errors in production,
   file the ones that aren't on the board with impact and a likely
   cause, and open a pull request for any one-line fix.
   
   Review every pull request marked ready for review against the rules
   in AGENTS.md. Comment only; never approve or merge.
   
   Every Friday, write release notes from this week's merged pull
   requests: what changed for users, grouped by area, with links.
   
   Take the next issue labelled "good first issue", work in your own
   git worktree, and open a pull request with tests.

How to set up AI engineers on Claude Code or Codex

1. Write the rules once

   Put build and test commands, conventions and review rules in CLAUDE.md or AGENTS.md, so every session starts from the same place. See CLAUDE.md and AGENTS.md for Codex.

2. Give each agent its own worktree

   Two agents editing one checkout overwrite each other. Use a git worktree per agent; see Claude Code worktrees and running multiple Claude Code agents.

3. Connect the tools the jobs need

   GitHub through its CLI, Sentry through its hosted MCP server, your tracker through its MCP server or CLI. Prefer CLIs where both exist; they use fewer tokens.

4. Keep merging with people

   Agents open pull requests and comment; people approve and merge. When an employee needs a decision, it marks the issue Needs action and asks.

What to expect

• Small, well-described issues land best: a bug with a stack trace, a missing test, a copy change across many files.
• Large features still need a person to break them into issues; agents are good at the pieces, less so at the plan.
• Usage comes from your own Claude or ChatGPT plan; schedule the heavy jobs inside working hours or overnight within your limits.
• Review load moves, it doesn't vanish: more pull requests means more to review, which is why AI review is on this list.

Engineers with a board, working hours and a heartbeat

Crewmark turns your own Claude Code or Codex into AI employees: each has a role, a goal in numbers, working hours and a heartbeat, takes the next issue from a shared board, comments as it works and opens pull requests for review.

1. Hire the role

   Start from the AI software engineer.

2. Add standing jobs as recurring issues

   Triage, review and release notes come back on their own schedule.

3. See other teams

   Marketing teams use the same setup: Claude Code for marketing. All jobs are on use cases.

FAQ