Knob

Model choice

You might say: model routing, capability tier

The policy that routes tasks to different model categories or tiers by difficulty and cost. Work that has become routine can run as plain code with no model.

What it does
Matches model capability and spend to the task and its risk.
Ask
Which model can finish this task reliably, and what evidence justifies the cost?
Typical
Smaller, cost-efficient models for easy or common work; more capable models for hard or unusual work; no model for work already turned into code.
Trouble
A poor route sends distinct task categories to one model, wasting cost or reducing performance.

What to look for

  1. A model name or tier in the run record
  2. A routing rule that sends easy and hard tasks to different models
  3. Cost and latency split by model choice

Connections

Sources

  1. Anthropic, Building effective AI agents, Workflow: RoutingRouting easy/common questions to smaller, cost-efficient models like Claude Haiku 4.5 and hard/unusual questions to more capable models like Claude Sonnet 4.5 to optimize for best performance.
  2. Kody, Documentation, About KodyYour connected agent does the thinking; Kody holds the durable result so the work can run again without another model in the loop.

Use it on your project

teardown-apply: hold "Model choice" (model-choice) 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