Task Decomposition & Routing Strategies

Task decomposition determines how complex work is broken into manageable units. The right strategy depends on whether the workflow is predictable or open-ended.

Fixed Sequential Pipelines (Prompt Chaining)

Best for predictable, multi-aspect workflows:

• Analyze each file individually, then run a cross-file integration pass
• Break reviews into sequential steps with defined inputs/outputs
• Each step's output feeds the next step's input

Example: Split large code reviews into per-file local analysis passes plus a separate cross-file integration pass to avoid attention dilution.

Dynamic Adaptive Decomposition

Best for open-ended investigation tasks:

• Generate subtasks based on what is discovered at each step
• First map structure, identify high-impact areas, then create a prioritized plan
• Plan adapts as dependencies are discovered

Example: "Add comprehensive tests to a legacy codebase" -- first map the codebase structure, identify high-impact untested areas, then create a test plan that adapts as you discover dependencies.

Routing by Complexity

Route tasks to the cheapest model that meets quality requirements:

• Use Haiku as a classifier to determine routing for more expensive models
• Simple tasks go to faster/cheaper models; complex ones escalate
• Monitor quality metrics per tier to validate routing decisions