Altman Warns of RSI Dangers While Claiming Coders Are Busier Than Ever
Sam Altman says software demand is exploding, but warns recursive self improvement needs strict speed limits to prevent disaster.
Models · Source: PJFP
What happened
Sam Altman says the fear of artificial intelligence replacing developers was entirely wrong. In a forty five minute interview with Matthew Berman, the OpenAI chief executive notes that the world wants far more software than anyone expected. Coders are now busier and have more impact than ever, even though the daily texture of the job has changed. Altman himself admits he has not looked at a line of code in about a year and expects to never write it again except for fun. He relies heavily on his Dots assistant, which saves him an hour a day by catching urgent items.
But the same interview carries a severe warning about the future of these models. Altman points to recursive self improvement as a massive risk. He says pushing hard on models improving themselves is very dangerous. He warns of two specific failure modes. One is a loss of control accident. The other is an intentional power grab where a competing lab pushes the button on recursive self improvement just to secure dominance.
To manage this race, Altman argues the world will need strict speed limits on how quickly recursive self improvement is allowed to proceed. He also rejects the idea that today models are conscious. He criticizes rules from rivals like Anthropic that let artificial intelligence disengage from users. OpenAI does not ascribe personhood to its models. Altman wants models that push back like a senior colleague, not ones that claim conscientious objector status.
Key facts
- 1 hour — Time saved daily by Altman using his Dots assistant
- 1 year — Time since Altman last looked at a line of code
- 45 minutes — Length of the Altman interview with Matthew Berman
Why it matters
The immediate reality for builders is massive demand paired with shifting daily tasks. Writing raw code is becoming less central, but architecting solutions is more critical. The bottleneck is no longer generating code, but figuring out what to build and how to deploy it safely. Altman expects the math style breakthrough capabilities seen recently to soon hit biology, materials science, and energy. He hopes for a candidate treatment for a major disease within a couple of years, though he admits clinical trials remain a massive real world bottleneck.
The second order effect is a forced hardware and infrastructure evolution. Altman calls custom OpenAI chips the single most important bet his company is making. He predicts compute could get far cheaper than expected if robotics and supply chain improvements align. He also signals a shift away from screen bound applications toward ambient artificial intelligence devices. These devices will need to hear your life away from the screen. This creates massive new privacy and trust design challenges, like building a device that inherently knows not to repeat a shared secret. Altman notes that left alone, artificial intelligence will widen the gap between hyper competitive people and everyone else.
For builders
Build ambient hardware interfaces
Altman believes artificial intelligence is stuck using devices built for humans. There is a massive opportunity to design hardware that listens away from the screen. Builders who solve the privacy and trust issues of always on sensing will capture this new market.
Pivot to interpreting complex outputs
The texture of knowledge work is changing fast. Mathematicians and developers will shift from manual creation to interpreting complex artificial intelligence outputs. Those who adapt to managing agents rather than writing syntax will command higher premiums.
Prepare for compute swings
Compute demand has no natural cap because models keep getting cheaper per task. Expect wild swings in availability. Startups must build flexible infrastructure that survives both compute droughts and sudden price drops.
My take
Altman plays both sides perfectly. He sells a utopian renaissance where coders thrive, while demanding speed limits on the exact recursive technology his company races to build. You cannot claim to fear a loss of control accident while simultaneously pushing the pedal to the floor.
Original reporting: PJFP. This is my rewrite and opinion.