PlanetArchitecture
Back to AI Fluency
L3

Delegator

Plan first, execute, review at end

Characteristics

Approach

Plan first, execute, review at end

Turns per task

2-5

Mental model

AI as capable peer you can delegate to

Error response

Review plan/approach, not just symptoms

Impact

Requests per feature

2-5 requests per feature

Rework rate

Low rework rate (15-20%)

Context quality

Clean context, focused execution

What It Looks Like

User: "Add email validation to user registration. Use writing-plans 
       to create a plan first."
AI: *invokes writing-plans*
    *creates comprehensive plan*
    "Here's the plan. Ready to execute?"
User: "Approved"
AI: *invokes executing-plans*
    *implements with checkpoints*
    *invokes verification-before-completion*
    "Done. All tests pass. Here's the summary."

Common Patterns

  • Always starts with planning
  • Reviews at checkpoints, not every turn
  • Uses skills consistently
  • Trusts AI to handle details
  • Intervenes only when needed

How to Level Up

The shift

Stop doing one thing at a time.

Practice
  • 1.Ask: "Are there independent tasks here?"
  • 2.If yes, use dispatching-parallel-agents
  • 3.Keep main context clean with subagent-driven-development
  • 4.Use async patterns for long-running work
Skills to learn
dispatching-parallel-agents — parallel worksubagent-driven-development — context managementusing-git-worktrees — isolated workspaces
Measure success

Did your turns drop from 5 to 2?