Dynamic Workflows Land in Claude Code

Anthropic introduced Dynamic Workflows in Claude Code on May 28, enabling runtime-flexible orchestration of multi-step coding tasks rather than the previously more static plan-then-execute model. The Hacker News thread (140 pts, 113 comments) was uniformly enthusiastic, with developers framing it as Anthropic's deepest investment yet in agentic coding tooling.

The feature pairs naturally with Opus 4.8's parallel subagent orchestration: Dynamic Workflows lets the model re-plan as it discovers state, while subagents fan out subtasks in parallel. Boris Cherny (Claude Code lead) has been publicly active on X discussing the broader sandboxing and permissions model for agents, signaling Anthropic's view that the next bottleneck is workflow control, not raw capability.

Competitive frame: this is Anthropic's direct response to Cursor's agent mode, GitHub Copilot Workspaces, and the new Mistral Vibe Work/Code modes. The same week, a viral Show HN ('Continue? Y/N' — a 60-second game about agent permission fatigue, 252 pts) captured the broader developer frustration that Dynamic Workflows tries to address.