In 2021, being a good software engineer felt great. The world was full of software, with more companies arriving every year who needed to employ engineers to write their code and run their systems. I knew I was good at it, and I knew I could keep doing it for as long as I wanted to. The work I loved would not run out.

In 2026, I’m not sure the software engineering industry will survive another decade. If it does, I’m certain it’s going to change far more than it did in the last two decades. Maybe I’ll figure out a way to carve out a lucrative niche supervising AI agents, or maybe I’ll have to leave the industry entirely. Either way, the work I loved is going away.

Tasting our own medicine

It’s unseemly to grieve too much over it, for two reasons. First, the whole point of being a good software engineer in the 2010s was that code provided enough leverage to automate away other jobs. That’s why programming was (and still is) such a lucrative profession. The fact that we’re automating away our own industry is probably some kind of cosmic justice. But I think any working software engineer today is worrying about this question: what will be left for me to do, once AI agents have fully diffused into the industry?

The other reason it’s unseemly is that I’m probably going to be one of the last to go. As a staff engineer, my work has looked kind of like supervising AI agents since before AI agents were a thing: I spend much of my job communicating in human language to other engineers, making sure they’re on the right track, and so on. Junior and mid-level engineers will suffer before I do. Why hire a group of engineers to “be the hands” of a handful of very senior folks when you can rent instances of Claude Opus 4.6 for a fraction of the price?

Overshooting and undershooting

I think my next ten years are going to be dominated by one question: will the tech industry overshoot or undershoot the capabilities of AI agents?

If tech companies undershoot - continuing to hire engineers long after AI agents are capable of replacing them - then at least I’ll hold onto my job for longer. Still, “my job” will increasingly mean “supervising groups of AI agents”. I’ll spend more time reviewing code than I do writing it, and more time reading model outputs than my actual codebase.

If tech companies tend to overshoot, it’s going to get a lot weirder, but I might actually have a better position in the medium term. In this world, tech companies collectively realize that they’ve stopped hiring too soon, and must scramble to get enough technical talent to manage their sprawling AI-generated codebases. As the market for juniors dries up, the total number of experienced senior and staff engineers will stagnate, driving up the demand for my labor (until the models get good enough to replace me entirely).

Am I being too pessimistic?

Of course, the software engineering industry has looked like it was dying in the past. High-level programming languages were supposed to let non-technical people write computer code. Outsourcing was supposed to kill demand for software engineers in high-cost-of-living countries. None of those prophecies of doom came true. However, I don’t think that’s much comfort. Industries do die when they’re made obsolete by technology. Eventually a crisis will come along that the industry can’t just ride out.

The most optimistic position is probably that somehow demand for software engineers increases, because the total amount of software rises so rapidly, even though you now need fewer engineers per line of software. This is widely referred to as the Jevons effect. Along these lines, I see some engineers saying things like “I’ll always have a job cleaning up this AI-generated code”.

I just don’t think that’s likely. AI agents can fix bugs and clean up code as well as they can write new code: that is, better than many engineers, and improving each month. Why would companies hire engineers to manage their AI-generated code instead of just throwing more and better AI at it?

If the Jevons effect is true, I think we would have to be hitting some kind of AI programming plateau where the tools are good enough to produce lots of code (we’re here already), but not quite good enough to maintain it. This is prima facie plausible. Every software engineer knows that maintaining code is harder than writing it. But unfortunately, I don’t think it’s true.

My personal experience of using AI tools is that they’re getting better and better at maintaining code. I’ve spent the last year or so asking almost every question I have about a codebase to an AI agent in parallel while I look for the answer myself, and I’ve seen them go from hopeless to “sometimes faster than me” to “usually faster than me and sometimes more insightful”.

Right now, there’s still plenty of room for a competent software engineer in the loop. But that room is shrinking. I don’t think there are any genuinely new capabilities that AI agents would need in order to take my job. They’d just have to get better and more reliable at doing the things they can already do. So it’s hard for me to believe that demand for software engineers is going to increase over time instead of decrease.

Final thoughts

It sucks. I miss feeling like my job was secure, and that my biggest career problems would be grappling with things like burnout: internal struggles, not external ones. That said, it’s a bit silly for software engineers to complain when the automation train finally catches up to them.

At least I’m happy that I recognized that the good times were good while I was still in them. Even when the end of zero-interest rates made the industry less cosy, I still felt very lucky to be a software engineer. Even now I’m in a better position than many of my peers, particularly those who are very junior to the industry.

And hey, maybe I’m wrong! At this point, I hope I’m wrong, and that there really is some je ne sais quoi human element required to deliver good software. But if not, I and my colleagues are going to have to find something else to do.