One strong model by default
Most objectives now run on one strong model from start to finish. Winskel adds planning, more models, and review only when the work earns them.
What changed
An objective now goes to one strong model that investigates, changes, tests, and repairs its own work in one session: Sonnet 5.5 when your Mac can run it, otherwise another strong model your accounts can run. It replaces routing every step by tier, introduced on September 22.
- Winskel plans a split with Claude Code when you ask for separate or parallel parts, or when the work involves security, authentication, migrations, infrastructure, or architecture.
- A split you ask for is kept. Independent parts run at the same time, and a dependent part starts on its own once the parts it needs passed their observed checks.
- Name an agent system for a part, such as "Use Codex for the tests", and that part stays on it.
Why it matters
Every handoff between models loses context and costs time. Keeping one owner on the work finishes plain objectives sooner, while larger or riskier objectives still get a plan and an independent review.
Limits
- A step whose checks fail moves once to a stronger model. After that, Winskel asks you.
- Claude Code is still required, because Winskel plans with it.
- Routing follows a fixed, versioned policy. It does not learn from your history.