All terms
Agents & tools
Cross-model handoff
Also known as: model handoff, context handoff, workflow handoff
Passing context, decisions, and intermediate work from one AI model to another — the connective tissue in a multi-model workflow.
What it means
A handoff is what happens between two stages in an AI team workflow. Model A finishes its part (planning, drafting, building), produces a structured artifact (a plan, a draft, a diff), and then model B picks it up. The structured artifact is the handoff document.
In practice, handoffs work best when they're explicit and structured: a PLAN.md, an IMPLEMENTATION.md, a REVIEW.md. Each one captures decisions, deviations, and open questions in a format the next model can read fresh. Without that structure, you end up dumping conversation history into a new context window and losing the signal in the noise.
The hard part is summarization. Model A often produces 100k tokens of conversation while solving a problem; model B doesn't need 100k tokens, it needs the conclusion plus the constraints that produced it. A good handoff brief is small (under 2k tokens), structured, and focused on what model B actually needs to make its next decision.
Example
Claude Opus writes a system design as PLAN.md (300 lines: goals, constraints, chosen approach, rejected alternatives). Claude Code reads PLAN.md and implements. Then Codex reads PLAN.md + the diff and reviews. Each model gets the structured artifact, not the previous conversation.
Why it matters
The biggest practical failure of multi-model workflows is bad handoffs. People paste raw chat history into a new model and wonder why output drops. Treating handoffs as first-class artifacts — written deliberately, kept short, focused on decisions — is what makes the team-of-models pattern actually work in production.