Five Claude agents in an afternoon? Commenters: the sixth one is for checking the first five

An Anthropic engineer made a claim at the Code w/ Claude event: you can build five Claude agents in one afternoon (Anthropic's AI agents, which can call tools to perform multi-step tasks), and each one can take over work you do manually every day.

The tweet includes a recording of a live talk. The speaker, William Steuk, presented on “Tool, skill, or subagent? Decomposing an agent that outgrew its prompt”: when an agent grows beyond what its prompt can carry, should you break it into a tool, a skill, or a subagent? The title itself says: building an agent is only the beginning; maintenance is the bigger part.

The tweet was quickly retweeted, and the comments section lit up.

One commenter worried: “Building five in an afternoon also means having five things you need to watch every day, when they go wrong.”

Someone used themselves as an example: “Five in an afternoon? My first agent took the whole weekend.”

The most brutal comment: “The sixth agent is the one you use to check the quality of the first five’s work.”

Another commenter noticed that the camera work looked like it was shot by a professional videographer. The production quality of the event video was not low.

Others said there is a lighter-weight approach: using @bot, you can make an agent in a few minutes, and they now have more than 20 running.

“Five agents in one afternoon” sounds like maxed-out efficiency. The comments section shifts the topic from build speed to another question: after they’re running, who provides the safety net? The video title already points in a direction: “Tool, skill, or subagent?” When an agent becomes too complex for its prompt to carry, it has to be broken up. But the precondition for breaking it up is that it has already grown to the point where it needs governance. Who watches the five agents? The answer from the comments: the sixth agent. It sounds like a joke, but people are already discussing it seriously.