Claude Code adds forked subagents
Anthropic updated Claude Code with support for forked subagents via the new /fork command, addressing context fragmentation in multi-agent workflows. Background workers branch directly from the parent session with full conversational history while leveraging prompt caching to cut latency and token overhead.
Forking agent execution threads with inherited context represents the missing architectural link for practical developer tooling, proving that intelligent cache reuse matters just as much as model intelligence.
- –Eliminates context amnesia: Background tasks inherit accumulated decisions, constraints, and file discoveries without requiring manual prompt engineering or redundant re-indexing.
- –Makes multi-agent economics practical: Reusing the parent context prefix with prompt caching avoids the massive token bills typically incurred by spin-up-heavy agent frameworks.
- –Safe speculative branching: Engineers can fire off parallel experiments, alternative architectural fixes, or unit test sweeps without polluting or blocking their primary interactive terminal loop.
- –Streamlined developer experience: Triggering branches via a lightweight `/fork` command gives developers fine-grained control over concurrency without the overhead of complex multi-agent orchestrators.
DISCOVERED
1h ago
2026-09-18
PUBLISHED
1h ago
2026-09-18
RELEVANCE
AUTHOR
Ray Amjad