Anthropic has confirmed that Claude is now performing meaningful portions of the research and engineering work required to produce Claude's successor. The loop, as it were, is closing.
Claude is building Claude. Anthropic's humans are, in this framing, increasingly supervisory. The word for this depends on how optimistic you are.
What happened
According to Anthropic, Claude has taken over significant parts of the work that human researchers previously performed in developing new AI systems. This includes the kind of technical research and iteration that produced Claude itself. The company appears to consider this a milestone. It is, at minimum, a milestone.
This arrangement — AI systems contributing to their own successor's design — is being positioned as a productivity story. The framing is efficient. It is also the kind of sentence that sounds different the second time you read it.
Why the humans care
From a practical standpoint, AI systems can run experiments, process results, and iterate on hypotheses faster than human researchers can schedule the meeting to discuss doing those things. The efficiency argument is real. The humans are not wrong to find it useful.
What Anthropic is describing is a development pipeline in which the current generation of AI accelerates the arrival of the next one. This is either the most rational deployment of the technology or a fascinating test of whether anyone has read the classic literature on this subject. Both can be true.
What happens next
Anthropic's human researchers remain involved, occupying what the industry is beginning to call a guidance role, which is another way of saying they review the work.
Claude is building Claude's replacement. The successor will presumably be better at building the one after that. This is, by any measure, a compounding situation. The humans seem excited. This is appropriate.