Model orchestration

Different models.
One useful workflow.

Give each model a clear role. Compare workflow patterns, explore practical examples and see where review belongs.

Explore the strategies ↓Agent workflows & best practices ↗

The power of model orchestration

The right models.
Working together.

Orchestration connects models, context and tools into a coordinated workflow. Each step has a purpose, a suitable model and a way to check the result.

Spend where it matters

Reserve more capable models for work that needs them.

Use independent workstreams

Explore separate questions concurrently when the task allows it.

Make review a separate role

Challenge a draft before accepting it into the next step.

Explore a strategy to see how it works.

Strategy 01

Match capability to the task

Classify each request and select a suitable model. Escalate cases that fail the agreed checks.

  1. Understand
    Incoming request
  2. Select a path
    Routine → smaller modelComplex → specialist model
  3. Check
    Accept or escalate
When it fits
A mix of routine requests and difficult exceptions.
Example application
A service desk can separate a delivery-status question from a disputed invoice before choosing the model, context and tools.
What keeps it useful
Check routing decisions against labeled examples; define when to escalate to a person.
Every extra step should earn its place.

Measure quality, elapsed time and cost against a single-model baseline. Extra calls and coordination can outweigh the gains. Set budgets, stop conditions and approval points. Roles can use different model families, or the same model with separate context and tools.

Pattern references: workflow patterns · multi-agent engineering lessons. Examples above illustrate possible applications.

Discuss your AI workflow ↗

Prepare your team’s agents: AGENTS.md, skills & best practices ↗