Lemire: after coding, agents will extend to running experiments and learning
lemire · x · 2026-08-28
Daniel Lemire argues that most professional programmers already build software with agentic AI, so the obvious next step is extending agents from software to hardware. Robotic chemists already exist, and in many cases this extension is well underway. The goal isn't AI-run factories — automated factories already exist — but AI that can experiment, learn, and draw lessons. Like coding, it will generate slop, but out of the slop might come nice surprises.
More from AGI Musings
- 2030s: The decade scarcity loses, driven by superintelligence, cheap energy, and automation — Dr_Singularity · 2026-08-28
- The Trap of AI Writing Assistants: You Must Ignore Half Their Suggestions — TuhinChakr · 2026-08-28
- AI Was Supposed to Give Us Free Time—Instead It 10x'd Output and 20x'd Expectations — djcows · 2026-08-28
- Can sustained interaction between machine minds bootstrap consciousness? Philosophers debate — AnnaCiaunica · 2026-08-28
- Real-World AI Core: You Must Become Part of the Machine to Teach It — yunta_tsai · 2026-08-28
- AI overlords and the end of software bloat — eyeronik1 · 2026-08-28