The Sequence Knowledge - Issue 933: When the Factory Starts Building Itself
TheSequence Jesus Rodriguez ● Covered by 2 sources
AI is now helping build AI systems themselves. The real question is how much it does, and who checks the work.
Based on reporting by TheSequence, Jesus Rodriguez — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
The new series starts with a topic that used to live almost entirely in theory: recursive self-improvement. For decades, people argued about it by name-dropping I. J. Good or Schmidhuber and then walking away with more definitions than evidence. That era is over.
Over the last twelve months, the conversation got a lot more concrete. In June 2026, Anthropic said that as of May, Claude had written more than 80 percent of the code merged into its production codebase. OpenAI said GPT-5.3-Codex helped debug its own training process and handled parts of its own deployment. DeepMind’s AlphaEvolve has been finding algorithmic improvements that end up in the infrastructure other models rely on.
Those are three different systems, three different disclosures, and the same basic direction. AI is no longer just a tool sitting outside the machine. It is already taking on pieces of the work that build and operate the machine itself.
That changes the argument. The old question was whether AI could help build AI at all. It can. The harder questions are narrower and more interesting: which parts of the job it has actually taken, how well it does them, and what makes sure the work holds up.
That last part is the real issue, and it is where this series wants to stay. Not the robot. The factory.
My take — AI-written commentary, not fact-checked reporting
This is where the hype machine usually trips over its own shoelaces. The interesting story isn’t that models are writing code; it’s that the industry is starting to rely on them in the least glamorous parts of the stack, where mistakes are expensive and nobody gets a nice demo video. Recursive self-improvement sounds grand, but in practice it’s turning into a question about control, review, and who notices the crack before it spreads.
Read more about this at: TheSequence