1,000 Claude Agents Proved Fermat's Last Theorem in 28.7 Hours — With No Boss and No Special Tools
Anthropic gave 1,000 AI agents one of the hardest problems in mathematics — a formal,...
Tag archive
Anthropic gave 1,000 AI agents one of the hardest problems in mathematics — a formal,...
A replay of our AI coding sessions running side by side: managers handing work to workers in parallel. They don't collide because every package has its own repository.

I measured how a client's request degrades as it passes through a team of agents, from lead to implementer. Names, numbers and prohibitions survive. The reasons behind them do not, and they are not worn away hop by hop: they are cut from the message to whoever builds. In prose, one model also drops the data contract, and its code stops working.
Why one agent with a real job title gets more done than a team of bots, and how to set yours up step by step.

Parallel agents are solved. Collaborative agents are not. The shared-session primitive is the gap everyone is racing to fill.
Livelock is the multi-agent failure that looks like diligence. A critic flags something, an executor fixes it, the fix trips a different check, and the crew spends real money making no net progress while every individual step looks correct. Why a per-agent loop cap cannot see it, the three shapes it takes (oscillation, escalation spiral, and the polite deferral loop), how to detect it from a hash of shared state instead of from the transcript, and a crew-level budget that turns an endless loop into a finite, reportable outcome.
The last article in this series ended with a count that was still zero. This one begins with a...
Article 2 in this series ended on the Agent Commons' hardest number. This article is about the...
If you run an agent — a research assistant with tools, a coding agent with a browsing habit, anything...
The Problem with Single-Model Workflows When you ask one large language model to design,...
GitHub's Big Bet on Multi-Agent Software Development In a significant overhaul of its Copilot...

In 2024–2025, enterprise AI meant prompt-chaining one giant model. That broke under multi-day...