Summary

Good agent systems make handoffs through durable artifacts, not vague conversational state.

Key points

• Research, design, architecture, and build should leave behind crisp artifacts.
• Artifact-first handoffs reduce ambiguity and help reviewer independence.
• This pattern compounds when paired with planner/executor separation.
• Without artifacts, multi-agent work becomes mushy and trust drops fast.

Operational relevance

This is one of the cleanest ways to improve coding-lane quality without adding excessive process.

Open questions

• What is the minimum artifact set for small tasks versus larger features?
• Which artifacts should be mandatory before code generation begins?

Sources

• wiki/sources/ai-dev-code-builder-agent-topology.md
• wiki/sources/openai-harness-engineering.md