Some tasks just don’t fit into a single conversation. An audit across the entire codebase. A migration spanning hundreds of thousands of lines. A research question that needs cross-checking. That’s exactly what Anthropic built Dynamic Workflows for — and with the latest Claude Code version, they’re now firmly anchored in the CLI.
What Dynamic Workflows are
The idea is simple and a little wild at the same time: you hand Claude a big task, and Claude writes itself an orchestration script for it. It plans the work, spawns dozens to hundreds of subagents in parallel in the background, lets them work on sub-problems — and then verifies its own outputs before reporting back. You manage all of it through the new /workflows command.
Put differently: Claude coordinates a small team made of itself. With Opus 4.8, those agents can run longer, which makes codebase-scale migrations possible from kickoff to merge — with the existing test suite as the bar.
Who can use it
Dynamic Workflows are available in Claude Code on the Enterprise, Team, and Max plans. On top of that, Fast Mode now defaults to Opus 4.8, at twice the standard rate for about 2.5x the speed. So if you want to push big jobs through, you get speed and parallelism at the same time.
My take
The feature launched in late May with Opus 4.8, but only now does it feel like it has truly arrived in daily CLI use. And it changes how you think about Claude Code. Until now the question was: how do I break my task down so it fits the context? Now the question is: how big am I even allowed to think?
Honestly: hundreds of agents in parallel sounds like a great demo moment and a hefty token bill. The real test is whether that self-verification at the end actually holds up — or whether you end up hand-checking hundreds of agent outputs afterward. If Anthropic nails this, it’s a real leap. I’m going to try it on a migration I’ve been putting off for weeks.