Claude now decides how to split work across agents

Anthropic used a workshop on the opening day of the AI Engineer New York conference to walk developers through dynamic workflows, a feature of Claude Managed Agents announced the previous Friday. The presenters, CJ Avilla and Harrison Stall of Anthropic's technical staff, kept calling them flows and described them as the next version of multi-agent orchestration.

The change is about who designs the team. Until now, a developer using Managed Agents configured a coordinator and a roster of subagents by hand. With flows, a lead agent drafts a plan and Claude then writes an orchestration program that launches agents, gathers their results and controls how information moves between them. Developers can still hand it preconfigured specialist subagents, and the whole thing runs in Claude Console.

The demonstration was modest on purpose: a stock watchlist scanner. A scan phase gave each ticker to its own agent with an isolated context, and a review phase checked the findings before a report was written. Avilla said the flow cost much more than a single agent doing the same job, but was much more effective at finding the details.

Audience questions marked out the current limits:

• How many agents a task gets is settled by evals and conservative defaults; ambitious jobs need to ask explicitly for more parallel work.
• The model does not know its budget while it runs. Developers set a session budget instead.
• Subagents share a filesystem and can exchange information through it, but the API offers no way to tell one subagent to message another.
• Every hand-off is inspectable: Console shows a graph with the payload the lead agent passed at each transition.

For making repeated flows cheaper, Avilla pointed to dreaming, a research-preview feature that reads a memory store alongside past session transcripts and produces a reorganised store, merging duplicates and replacing stale entries.