Parallel Agents

Running multiple agents can help, but only when the work is clearly separated.

The working rule

One agent, one bounded task, one clear ownership area.

Good uses

• one agent on docs
• one agent on checkout QA
• one agent on a protected-screen implementation

Bad uses

• two agents editing the same files
• vague “go figure it out” work
• running parallel agents when the next step depends on one blocking answer

Practical advice

• avoid worktrees unless you truly need them
• keep file ownership clear
• merge or review one bounded result at a time
• use the main thread to coordinate, not to duplicate work

What success looks like

Parallel agents help when they increase throughput without creating merge conflict cleanup or duplicated investigation.