先说结论

If you mostly use Claude as a chat and coding assistant, and every big launch makes you wonder whether you are about to get more leverage or just more overhead, this is the part to watch. The easy mistake is to treat Claude Code and Claude like the same product with different scores. Do that, and what sounds like more power may first show up as stricter structure and more overhead.

The scene is familiar: you open the May 28 launch expecting the usual question, did the model get stronger?

Then you hit the real split. In Dynamic Workflows in Claude Code, the workflow is a JavaScript script Claude writes to decide the next step, while intermediate state lives in script variables instead of the chat thread.[S001] Workflow is not more agents. It is the plan leaving the chat and moving into code.[S001]

That matters because long tasks break differently in chat. You keep restating the plan, correcting drift, and rebuilding context. Once the plan lives in code, you can inspect it, rerun it, and revise it without re-explaining the whole job. The hidden cost of missing this is that you keep using Claude Code and Claude for the wrong jobs and wonder why the workflow feels messy.

为什么这次值得看

Anthropic's May 28, 2026 launch post made the scale obvious: Claude can dynamically write orchestration scripts and drive dozens to hundreds of parallel subagents.[S002] But I do not think the main story is the parallelism. The most discussable part is never that the model got stronger. It is why the strongest thing was not shipped as the default.

关键证据

My current read, based on the workflows docs and the launch post, is simple: Claude Code is better for helping you see the problem clearly first; Claude is better for carrying the rest of the work through. I would treat this as a planning feature first, not proof that every long task is now solved end to end.[S001][S002]

So the practical move is not 'use more agents.' It is: ask for the workflow itself, save the script once it works, and reuse it on the next repeatable task. If someone on your team still evaluates coding tools like a leaderboard, share this with them. What would you script first: repo audit, refactor planning, or code review?

适合谁 / 下一步怎么用

最后落到动作：share