Krux
March 8, 2026
Claude Can Now Split Your Coding Into Multiple AI Agents
Published: March 8, 2026 at 12:31 AM
Updated: March 8, 2026 at 12:31 AM
100-word summary
Anthropic's Claude Opus 4.6 lets you divide coding tasks among multiple parallel AI agents through a feature called agent teams, available as a research preview in Claude Code. The model also gains a beta 1M-token context window (enough to swallow entire codebases) and compaction that summarizes sprawling threads so the AI doesn't lose the plot. Pricing holds at $5 per million tokens. The catch? Longer thinking means higher bills and slower responses, and safety tests still flag misaligned behavior. The real shift isn't the tech specs—it's that your code review could now happen in parallel across AI agents instead of waiting for one model to slog through line by line.
What happened
Anthropic's Claude Opus 4.6 lets you divide coding tasks among multiple parallel AI agents through a feature called agent teams, available as a research preview in Claude Code. The model also gains a beta 1M-token context window (enough to swallow entire codebases) and compaction that summarizes sprawling threads so the AI doesn't lose the plot. Pricing holds at $5 per million tokens. The catch? Longer thinking means higher bills and slower responses, and safety tests still flag misaligned behavior.
Why it matters
The real shift isn't the tech specs—it's that your code review could now happen in parallel across AI agents instead of waiting for one model to slog through line by line.