What is an agent fork?

Definition

An agent fork copies an established conversational or project context into a new branch of work. The fork can explore an alternative, handle a subtask, or continue separately without changing the original task's subsequent conversation.

Forked work needs a clear starting state and integration plan. Concurrent branches can diverge in assumptions or modify overlapping files, so their results should be compared and reconciled deliberately before reuse.

ELI5

An agent fork creates a separate AI-agent task from the completed history of an existing task. The new branch can explore another approach without changing the original task's later conversation.

For example, one fork can try a minimal software fix while another explores a larger redesign. Their assumptions and file changes may diverge, so people must compare and reconcile the results before combining them.

Acronyms and aliases

fork mode variant

Frequently asked questions

How is an agent fork different from a new task?

A fork begins with inherited completed history, while a new task normally starts from only the prompt and explicitly supplied context.

Can two agent forks safely edit the same files?

They can, but overlapping changes create conflicts and require careful review, ownership boundaries, and deliberate integration.

Videos explaining agent fork