Concept

Parallel runs

You might say: parallel agents, background agents, fan-out, coordinator and subagents

Several runs work at once, each in its own worktree or runner, so they do not fight over files, ports or build output. A coordinator run can split a job, hand the parts to subagents and write no code itself. Each subagent works in its own context window and reports back one result. Merges still happen one at a time. Each extra run adds a pull request to the review queue, so parallel runs pay off only once verification carries most of the review.

What it does
Multiplies how much work the factory attempts at once, while each attempt stays isolated.
Ask
Does each run get its own worktree or runner, and who will review everything they produce?

What to look for

  1. Several runs in progress at once on the run page
  2. One worktree and branch per run
  3. A coordinator run that hands out briefs and opens no pull request of its own

Pattern

Kitchen brigade

Connections

Sources

  1. Anthropic, Run parallel sessions with worktrees, Run parallel sessions with worktreesA git worktree is a separate working directory with its own files and branch
  2. Anthropic, Run parallel sessions with worktrees, Run parallel sessions with worktreesSubagents can run in their own worktrees so parallel edits don't conflict.
  3. Lauren Tan, pstack, Give parallel work its own machineIf you run several agents against one repository on one computer, they will fight over the working tree, the ports, and the build output.
  4. Lauren Tan, pstack, Steer with principle namesseparate before serializing shared state. give each attempt its own worktree, no locks.
  5. Lauren Tan, pstack, Run many projects in parallelThe coordinator doesn’t write code. It directs subagents, which run in the cloud by default
  6. Lauren Tan, Noodle, docs/concepts/scheduling.mdMultiple agents run in parallel, each in its own git worktree.
  7. Lauren Tan, Noodle, docs/concepts/scheduling.mdMerges are serialized, one at a time, to keep the main branch clean.
  8. Lauren Tan, Noodle, INSTALL.mda head chef (the scheduler) reads the backlog, writes orders, and dispatches line cooks (agents) to do the work in isolated worktrees.
  9. Matt Pocock, AI Coding Dictionary, SubagentRuns in its own session with its own context window, and reports a single tool result back.
  10. Matt Pocock, AI Coding Dictionary, Subagenta parent can fan several out at once over independent pieces of work.
  11. Anthropic, How we built our multi-agent research system, How we built our multi-agent research systemOur Research system uses a multi-agent architecture with an orchestrator-worker pattern, where a lead agent coordinates the process while delegating to specialized subagents that operate in parallel.
  12. Mohannad Hammadeh et al. (AWS), Scaling agentic coding past the human bottleneck, Scaling agentic coding past the human bottleneckeach agent you add makes one review queue longer.

Use it on your project

teardown-apply: hold "Parallel runs" (parallel-runs) from the teardown at https://teardownatlas.com/software-factory/ against my project

Paste it into a coding agent that has the teardown skills.

An AI agent wrote this map with the teardown skill. It can contain errors: check the sources before you rely on a part. About and legal