Why software engineers will survive the AI era
I am skeptical of the superintelligence narrative.
Software engineers will only be fully replaced if the “fast takeoff” AI narrative that the AI labs and AI boosters are pushing comes true.
I spend a lot of time in AI communities that believe it will, and lately they’ve had me panicking.
They are certain it’s coming, and the consequences are enormous. But lately, I’ve been much more skeptical of their claims.
If you aren’t familiar, fast takeoff is the idea that AI is improving exponentially at everything and that superintelligence (SI) is right around the corner. The dates vary, but many people point to 2027 because of the famous AI 2027 paper.
When they talk about SI, they mean AI so intelligent it can beat a human at any intellectual task.
This isn’t a fringe view either. Anthropic CEO Dario Amodei, OpenAI CEO Sam Altman, and Elon Musk have all predicted AI that surpasses humans within the next few years.
Here is what they think will happen.
First comes RSI: recursive self-improvement. That’s when AI becomes so good it can improve itself without any human intervention.
RSI makes AI progress accelerate even faster than it does now, which leads directly to superintelligence.
It doesn’t end there, though. The next step is putting SI into robots and building hundreds of millions of them.
Once there are hundreds of millions of superintelligent robots, human labor is over. AI will do every white-collar and blue-collar job better than we can.
The people cheering this on, the accelerationists, believe we’ll then reach a post-scarcity world where everything is abundant, the government pays us all a universal basic income, and nobody has to work.
Many people have concluded this is inevitable and that it starts next year.
When someone tells you AI will replace software engineers, this is the story they’re telling.
I’m skeptical.
The biggest hole I see is RSI. The accelerationists and AI companies act like RSI is basically here, and the companies have billions of dollars riding on you believing it. People selling shovels always say there’s a gold rush.
I’ve seen no compelling evidence that RSI exists outside of AI researchers using AI to help them do their jobs.
That’s not RSI; it’s a better tool.
RSI has never happened. Until it does, the burden of proof is on them.
What I see instead is AI improving exponentially at some tasks and linearly at others. It writes code faster every month, but it still can’t decide what’s worth building.
OpenAI’s GPT-6.2 Astra is the latest example. It launched this week with big benchmark gains, and early testers tore it apart for stalling and looping on long tasks.
To replace software engineers, AI needs judgment as good as a human’s. I see no evidence of exponential progress toward that, and I’m not changing my mind because AI labs keep hinting they might secretly have RSI.
AI will shrink engineering teams, and it’s already cutting junior hiring. But smaller teams aren’t the end of the profession.
AI will change what software engineers do. It won’t replace them.