Anthropic launches Dynamic Workflows for Claude Code, orchestrating hundreds of parallel subagents per session

Anthropic on May 28, 2026 introduced Dynamic Workflows in Claude Code, a research preview that lets Claude plan and execute long-horizon coding tasks by orchestrating tens to hundreds of subagents running in parallel within a single session. The feature shipped alongside the Claude Opus 4.8 launch and is available today in the Claude Code CLI, the Desktop app, and the VS Code extension for Max, Team, and Enterprise plan subscribers.

What's new

• Distribution. Dynamic Workflows are gated to Max, Team, and Enterprise plans, and exposed across all three Claude Code surfaces: "Dynamic workflows are available today in research preview in the Claude Code CLI, Desktop, and the VS code extension for Max, Team, and Enterprise plans."
• The trigger flow is conversational. "When a workflow kicks off, Claude plans dynamically based on your prompt, breaks it into subtasks, and fans the work out across subagents running in parallel."
• Script-level orchestration is the load-bearing piece. Behind the scenes, Claude "dynamically writes orchestration scripts that run tens to hundreds of parallel subagents in a single session, checking its work before anything reaches you." This is the first time Claude Code exposes script-level orchestration to end users instead of doing it in-conversation.
• Explicit cost surfacing. Anthropic flags the cost profile up front: "Dynamic workflows consume meaningfully more usage than a typical Claude Code session." The first time a workflow triggers, Claude Code shows what's about to run and prompts the user to confirm before any subagents fan out.

Context

Dynamic Workflows is the second half of Anthropic's May 28 Claude Opus 4.8 release wave, which also defaulted the API context window to 1M tokens, raised the effort parameter default to high, and brought Computer Use, the advisor tool, and Claude API Fast Mode onto the new flagship. The Workflows preview is also called out in the Claude Platform release notes as "available as a research preview, letting you define and run multi-step agentic plans" — confirming the surface is a first-party CLI/IDE capability rather than a third-party wrapper.

Pre-Workflows, scaling Claude Code to multi-hundred-step plans meant chaining sessions externally or leaning on community harnesses. Plan-and-execute behavior was already present, but execution was largely serial inside a single conversation, with Claude itself acting as the orchestrator. The new preview pushes orchestration into generated scripts that the model writes and runs, so a single prompt can spawn a tree of subagents working in parallel rather than a single linear chain.

Why it matters

Dynamic Workflows closes one of the most-cited gaps between Claude Code and competing agentic coding tools: that long-horizon planning and parallel execution had been left to users to wire up. By generating an orchestration script and fanning subagents across it inside a single session, Claude Code now has a first-party answer to the "long task fleet" pattern that GitHub Copilot's coding agent, Cursor's background agents, and OpenAI Codex's task tool have been advancing this year.

The framing is cautious by Anthropic standards. Gating to paid plans, calling it a research preview, surfacing cost expectations up front, and inserting an explicit pre-run confirmation step are all consistent with how Anthropic typically rolls out features that can rack up usage quickly. For teams already on Max, Team, or Enterprise, the practical effect is that Claude Code can now absorb a much larger atomic chunk of work — refactors across a whole repository, end-to-end feature builds, multi-file test scaffolding — without the user having to hand-decompose the plan. The direction of travel is clear: Anthropic is moving Claude Code from "smart pair programmer" toward "shippable autonomous engineering harness," and Dynamic Workflows is the load-bearing piece of that transition.