AI companies push users from chatting with bots to managing agent teams

Anthropic and OpenAI this week released products that recast AI from a single conversation partner into a supervised workforce of agents. The idea is that developers and knowledge workers will dispatch tasks, monitor parallel threads, and step in when an agent needs direction.

The approach is unproven: current agents still require heavy human intervention to catch errors, and no independent evaluation has shown that multi-agent setups reliably beat a single developer working alone. Anthropic paired a new flagship model, Claude Opus 4.6, with an "agent teams" feature in Claude Code that lets developers spin up multiple subagents to split work, coordinate autonomously, and run concurrently.

The teams appear as a split-screen terminal where a developer can jump between subagents, take over any one directly, and watch the others keep working.

anthropic, openai, claude opus, claude code, agent teams, subagents, multi-agent, concurrent agents, human intervention, developers