What if every software developer on earth was replaced by AI agents running 24/7?
And I mean including AI/ML researchers code too.
I mean ALL the code in Python, C++ etc. that causes the creation of our frontier models like Claude Opus 4.6 and Kimi K2.5, as well as their data generators for training etc.
Which means recursive improvements in AI.
I don’t know exactly how we get there. But enough sharp people are pushing hard enough that I believe it lands within 1–2 years.
And once it does, the loop closes on itself: better AI writes better code → better code accelerates AI research → that produces better AI. Software ecosystems start evolving on their own, then spill into robotics, factories, and hardware — all running at machine speed.
Dario Amodei calls coding “the first domino.” Faster code speeds up AI research, which produces faster code. He sees AI owning the full stack — architecture, debugging, deployment — with no human in the chain. His endgame: a country of geniuses in a data center, never sleeping.
After software falls, the rest of the economy follows fast. I’m betting that’s visible by 2027 but that’s just my humble guess.
As for what humans do — my guess is we become swarm masters. Not slowing things down, just injecting occasional signal: weird ideas, unexpected judgment, imperfect intuition fed back into an otherwise machine-driven loop.
A ghost in the machine, by choice.
Machine teaches human → human comes up with something weird → machine likes it. Repeat.
And I mean including AI/ML researchers code too.
I mean ALL the code in Python, C++ etc. that causes the creation of our frontier models like Claude Opus 4.6 and Kimi K2.5, as well as their data generators for training etc.
Which means recursive improvements in AI.
I don’t know exactly how we get there. But enough sharp people are pushing hard enough that I believe it lands within 1–2 years.
And once it does, the loop closes on itself: better AI writes better code → better code accelerates AI research → that produces better AI. Software ecosystems start evolving on their own, then spill into robotics, factories, and hardware — all running at machine speed.
Dario Amodei calls coding “the first domino.” Faster code speeds up AI research, which produces faster code. He sees AI owning the full stack — architecture, debugging, deployment — with no human in the chain. His endgame: a country of geniuses in a data center, never sleeping.
After software falls, the rest of the economy follows fast. I’m betting that’s visible by 2027 but that’s just my humble guess.
As for what humans do — my guess is we become swarm masters. Not slowing things down, just injecting occasional signal: weird ideas, unexpected judgment, imperfect intuition fed back into an otherwise machine-driven loop.
A ghost in the machine, by choice.
Machine teaches human → human comes up with something weird → machine likes it. Repeat.