I see two competing visions of how AI is going to affect the tech industry and, in particular, software engineers.
- This is going to be like what cars did to the horses and carriages.
- This is going to be like what offshoring did to software engineers in the US.
Of the two, the second seems like the better analogy to me. I still remember there was a famous author, Ed Yourdon, who wrote a book (The Decline and Fall of the American Programmer) in 1992(!), predicting that US software developers would mostly find themselves unemployed because of a coming wave of offshoring to India. He was right that offshoring to India would be a huge wave. He was wrong about the effects on US employment (and, to his credit, wrote a follow-up book three years later admitting that).
Now, in 2026, the AI models look suspiciously similar to the offshore developers of the 1990s. They are maybe not quite as good (on average) as the top level US developers, but they are much cheaper and willing to work hard to make up.
The scary prediction is that we’re early on an exponential curve of AI achievement, and AI will shortly be better than all human software engineers and, some time not too long after that, better than all white collar workers. I can’t say for sure that won’t happen, but I think that curve won’t hit us all that soon.
Most importantly, the AI models are getting better at an exponential rate, but in part that’s because the amount of money and other resources devoted to them is also growing at an exponential rate. The frontier labs earn respectable amounts of money, especially for companies that are so young. They still manage to lose fabulous amounts of money, growing each quarter, and they have no way, so far, to recoup that. Every AI product is a loss leader.
We’ve seen this before, of course, at least three times. Amazon famously didn’t make money on retail for more than a decade before it did. Google lost money as a search engine and seemed to be stuck because people trusted it partly because its search results were not polluted with ads. Then they added ads after all, they just avoided the most annoying and counter-productive ones, at least on their own site. (I’m talking about Google before about 2008.) The cloud providers, starting with AWS, didn’t make money in their early years and now they make so much money there’s a new name for them, “hyperscalers”.
The AI labs could do this, and perhaps only one of them needs to actually succeed, but there is also a future in which they don’t get AGI but they do find a business model that falls short of that and is still profitable. Maybe I’m blind because of my own stake in the game as a working software engineer, but that seems like a reasonably likely scenario. It rhymes with what happened with offshoring: lots of work went to India, but mostly the kind where cost was important and other things could be compromised on. (What Yourdon didn’t fully predict was that lots of Indian software developers would come to the US and become a vital part of the ecosystem there.)