Skip to content

AFK Implementation / Night Shift

matt-pocock's term for the phase where, once alignment and planning are curated, the human exits the loop and lets agents grind through a Kanban of vertical-slice issues in parallel.

"It's at this point that the human leaves the loop. … We can think of this as kind of like the day shift and the night shift. This is the day shift for the human, right? Planning everything, getting all the stuff ready. And then once we kick it over to the night shift, the AI can just work AFK." — Pocock, 52:26–53:34

Shape of the flow

  1. Day shift (human-reviewed): grilling → PRD → issue decomposition into vertical slices with blocking dependencies recorded on a Kanban board.
  2. Night shift (AFK): an agent — or a small fleet of agents — picks issues off the board, explores the repo, implements, runs TDD + type-check feedback loops, and produces reviewable commits.
  3. Morning: human QA merges what passed, files new issues for what didn't.

Sandcastle

Pocock's own harness for this phase. Key pieces (from the main.ts walkthrough around 2:10:00):

  • plan prompt selects N non-blocking issues from the backlog for this wave.
  • For each issue: create a git worktree, spin a Docker sandbox, run the implementer prompt against the issue.
  • If commits land, run a reviewer agent (with pushed coding standards) against the diff.
  • A merger agent consolidates green branches; resolves type/test conflicts.
  • Pocock uses Sonnet for implementation, Opus for review ("I need the smarts then").

Preconditions

AFK only works if the codebase has strong feedback loops, deep modules, and tightly scoped tracer-bullet issues — otherwise the night shift comes back to merge conflicts and dumb-zone code.