AI handles everything from planning to assigning tasks
Anthropic has added dynamic workflows to its AI agent execution platform, Claude Managed Agents, allowing multiple agents to divide up the work. A lead agent makes a plan, assigns tasks to subagents and integrates their results. Up to 1,000 agents can run in parallel in a single execution.
An AI agent is an AI system that works toward a given goal. With this new feature, the work is divided between a lead agent and agents that handle specific tasks, rather than leaving everything to one agent. In effect, a mechanism for running multiple AIs together has been added to the existing execution platform.
The limit of 1,000 is not the only thing worth noting. The AI itself handles dividing up the work and consolidating the results. However, the number of agents that can run in parallel does not directly translate into faster processing or proportionally better results.
Detected 66 of 70 bugs
In a test cited by Anthropic, 70 bugs were intentionally planted in 116,000 lines of code, and the results were compared. A single agent found 14–27 bugs in one run, while the dynamic workflow reportedly found 66 consistently.
That translates to a detection rate of about 20–39% for a single agent and about 94% for the dynamic workflow. In this test, having multiple agents divide up the work led to more bugs being detected. Even so, the result of 66 means that not all the planted bugs were found.
The result also comes from a specific task: detecting bugs in code. It is unclear whether the same difference would emerge in other types of work, such as research or document writing. The materials provided also do not say how many agents were actually used in the test, so it would be premature to assume the result was achieved with 1,000 agents.
Measure adoption by cost per result, not by the number of agents
Dividing up the work also comes with the burden of increased AI processing. Anthropic warns that the system may consume a large number of tokens and recommends starting with a small setup. Tokens are units of text or code processed by AI and serve as a measure when considering usage costs.
What businesses need to determine is not how many AI agents they can run, but how much it costs to obtain results of the required quality. Even if more bugs are found, adoption decisions will depend on how much costs and review work increase. A practical approach is to compare a single agent with a small multi-agent setup on the same task, measuring the number of bugs detected, time taken and cost.
The materials provided do not give the total cost, processing time or token usage for this comparative test. What can be confirmed is the difference in the number of bugs detected; the results do not demonstrate cost-effectiveness.
Comments