We're not losing our jobs

Every model release comes with the same announcement: software engineers are finished. It arrives from the people selling the models, or from the people whose relevance depends on being first to hype them. I've been hearing it for years now, and every time it lands I believe it a little less.

Part of what's changed is that the hysteria has cooled. The mass panic of the early years has tapered off, at least in my circles on X, and what's replaced it is a kind of collective exhaustion. AI has been permuted into every corner of everyday life and a lot of us are getting sick of it. I'm glad. The narrative around this technology was always more destructive than the technology itself.

I built a program using AI front to back that's been instrumental in running my Kubernetes cluster. At work they've been genuinely helpful for busy-body work — turning Slack messages into tickets, doing a quick high-level investigation off a ticket before anyone opens the code. For PR reviews and small mechanical tasks, they're good. On the micro scale, writing an individual function or a small piece of software, they can be excellent. That's a real category of value and I won't pretend otherwise.

But if you use these models every day, you also know how smart and how incredibly stupid they are at the same time. They don't follow rules. They hallucinate constantly. They will tell you with total confidence that what they're doing is correct, right up until they delete System32 and apologize.

I used to let them run wild — build the whole feature, and I'd check near the end that it worked. That's over. With every new model the behavior gets less predictable, and Opus 5 has been the clearest example. It over-engineered a feature I was working on to an absurd degree. I ended up nuking half of what it wrote. Whatever it's being touted as, that isn't the experience of using it. I'm now looking forward to lowering AI's priority in my workflow and getting back to writing code the normal way.

Meanwhile, companies are laying off entire departments on the theory that an LLM can cover the gap. I'd find that funnier if people weren't losing their jobs over it.

Which is why the role these things play matters so much. Models are good working alongside you. They should hold no position of power in your company or in your own process. Use them for PR reviews. Use them for the tedious stuff. Put them in CI if it helps. But never let AI be the final say, and never let AI be the go/no-go on a PR. (And let's not get started on what atrophies when you hand every bit of the effort over — that's a post for another time.)

All of which brings me back to the announcement.

Anyone telling you an LLM is about to take your job almost certainly stands to gain from you believing it. That's what happens when everybody is selling shovels. The same goes for the gatekeeping — the line that if you're not using LLMs every single day for everything, you should pack your bags and go farm potatoes. The same goes for the guy who one-shot a video game with Claude and wants you to know that the future of gaming has arrived. Notice how the demo is always a demo.

We should push back on all of it. Not because AI is bad, but because being told every other week that your career is ending is corrosive, and because the people saying it are either selling something or have talked themselves into believing their own pitch.

I'll admit I'm also here for the content. When this bubble finally implodes — this money-changing-hands, propped-up valuation built on nothing but smoke and mirrors — I intend to watch it go down in flames with a bag of popcorn in my lap.

There is still an enormous need for software engineers. That hasn't changed, and no model release has changed it. Make sure you have your popcorn ready.