On Thursday, Anthropic and OpenAI released products built around a new idea: instead of conversing with a single AI assistant, users would manage teams of AI agents that divide tasks and run in parallel. The launches mark a broader industry shift from AI as a chat partner to AI as a delegated workforce, arriving during a week when such an approach rattled software stocks.
Experts caution that the supervisory model remains unproven in practice. Even the most advanced multi-agent setups still require significant human oversight to catch mistakes, and there is no independent consensus that these tools outperform a lone developer.
Anthropic rolled out Claude Opus 4.6 paired with Claude Code’s “agent teams.” The feature lets developers spin up several AI agents that split a task, coordinate autonomously, and run concurrently. In use, agent teams resemble a split-screen terminal where a developer can switch among subagents and watch the others work. Anthropic says the tool is best for tasks such as codebase reviews and is available as a research preview.
OpenAI introduced Frontier, an enterprise platform that assigns each AI agent its own identity, permissions, and memory and connects to business systems like CRMs and data warehouses. Barret Zoph, OpenAI’s general manager for business-to-business, framed Frontier as a way to hire AI co-workers who perform many existing computer tasks, while the company emphasizes Frontier is not trying to replace all software but to complement it.
