Anthropic shipped Dynamic Workflows for Claude Code alongside Opus 4.8 last week, and this one’s worth paying attention to if you’re building anything at scale. The feature is in research preview for Enterprise, Team, and Max users — and it fundamentally changes what a single Claude session can do.
What Dynamic Workflows actually means
Instead of one Claude instance working through a task linearly, the model can now plan work, spin up hundreds of parallel subagents, verify their outputs, and consolidate results — all in a single session. Anthropic says Claude Code can carry out codebase-scale migrations across hundreds of thousands of lines from kickoff to merge, using the existing test suite as the bar for completion.
That’s not a demo use case. That’s the kind of task that usually means a week of coordination across a team.
The honesty improvement matters more than the benchmark
Opus 4.8 moved from 64.3 to 69.2 on SWE-bench Pro, but the change I find more significant is reliability: the model is four times less likely than 4.7 to let code flaws pass without flagging them. For agentic work where you’re not watching every step, that’s the property that actually matters.
If you’re on Enterprise or Max, the research preview is available now. Worth spinning up a migration task and seeing how far it gets before you need to intervene.