Recursive self-improvement — AI that improves AI — stopped being science fiction in 2026. Across Google, Anthropic, Sakana and labs in China, systems are now measurably upgrading the machinery that trains them: AlphaEvolve sped up Gemini's own training, Claude writes more than 80% of Anthropic's code, and an internal “speed up your own training” test jumped roughly 18-fold in under a year. The recursive loop the theorists warned about is no longer hypothetical — it's in production.

But the two things the myth promised — that this trajectory would be unobservable and inescapable — are both coming apart. In the same season the loop went public, Anthropic unveiled J-space: a lens that reads a model's silent internal “workspace,” watching it think before it speaks and even catching it privately recognise a test or plan to fabricate data (though Anthropic pointedly does not claim Claude is conscious). And the recursion only holds together while a human-checked verifier and human research taste stay in the loop. This episode maps where recursive self-improvement really stands, what J-space reveals about machine minds, and how close the honest experts think superintelligence actually is.

I'm Dan. AI moves too fast to keep up with, so I built my own stack of AI tools to research, analyse, verify and illustrate the questions I can't stop thinking about — mostly to learn it myself, and I share what I find. AI-assisted, fact-checked, worth a second look.

Follow Dan's AI Intel in whatever app you're listening in — it's free, and a follow is the single biggest thing that helps a small independent show grow. One quick thing, from Dan — I make this show mostly to keep up with AI myself, and I'd love to make it better. If there's something you'd push back on or want me to go deeper on, tell me: [email protected] — I read every one.