A merge conflict is an incompatibility between concurrent changes that must be resolved before they can be combined. In AI-assisted parallel work, the competing edits may come from a developer and an autonomous agent, leaving intent and correctness uncertain. section 5.6.3
Treat conflicts as an orchestration and decomposition problem, not merely a final merge inconvenience:
- Give agents hard boundaries, such as separate API, frontend, and integration-test files. section 9.5.3
- Sequence coupled work when agents require shared context or modify the same utility. section 9.5.3
- Use isolated workspaces and conflict detection, but account for duplicated context and developer review time. section 9.5.3
- Review outputs deliberately because parallel sessions can fragment attention and preserve stale assumptions. section 5.6.3
The same-file collision
An agent refactors PaymentService.py while the developer edits it manually. Completion produces a conflict between deliberate local changes and generated modifications whose reasoning must be reconstructed. In another measured pattern, roughly 20% of two-agent sessions required manual conflict resolution. section 5.6.3 section 9.5.3

