Editorial

Claude Managed Agents now lets an agent write the program that orchestrates up to a thousand other agents, run it at once, and keep its code out of view. That is how databases treat a query plan. A database can hide its plan because the plan cannot change the answer. This one decides what the answer is.

Dynamic workflows hide an agent's orchestration program the way a database hides a query plan, but a plan may be private only when it cannot change the result, and this one decides what is read, what is retried and what counts as finished.

On 2 October a merged Kubernetes design document was reverted out of the tree — 2,020 lines removed — because a maintainer concluded it had been written with undisclosed AI assistance. The project reserves a status for proposals that are kept as historical documents, and has nowhere in the repository to put the fact that decided this one.

Kubernetes requires AI disclosure in the pull request description and forbids the commit trailer that would carry it, so when provenance decided the fate of a merged design document, the only way the record could express doubt was to delete it.

The same commit trailer is required by the Linux kernel, required by Node.js, and forbidden by Kubernetes. Major projects are now writing their contribution rules down twice — once for people, once for the machine — and nothing checks that the two versions still agree.

Every serious project now keeps two copies of its contribution process, one addressed to contributors that carries the authority and one addressed to agents that carries the effect, and the second copy is a derived artefact whose derivation nothing enforces.