Those of you who aren’t in the tech industry may not realize the extent to which it’s... I’m trying to think of a word other than “disruption” — such a cliché — but it’s really the word that fits. Past transitions — the shift from server-side to client-side, the great XML delusion, the battle over the “semantic web” — feel like child’s play compared to what AI is doing (not to mention the amount of disagreement about how much it’s doing, and how fast).
If you scroll through the tech side of LinkedIn: AI is slop, we still review every line of code by hand; AI is so useful we’re laying off all the human engineers; it’s just another tool; it’s good for this but bad for that; it’s the end of the industry. There’s no agreement. That’s because the disruption is still happening, and like all change it’s happening unevenly.
I can’t relate to the “AI is useless” side. I’ve worked with AI as a co-developer on small personal projects and on decades-old legacy corporate codebases; I’ve found it to be scary good at both writing and refactoring code in both situations. And improving rapidly, on a scale of months, in its usefulness and trustworthiness.
(And this would make sense even for those who still believe AI is just a stochastic parrot, that it’s “just next-token prediction” and so can only repeat variations on what’s in its training data. At least for coding. Because like 90% of day-to-day coding work is repetition of existing techniques or algorithms. At the end of the day, what most of us are working on is just another database with a front end and with business rules tacked on. Intricate and delicate, sure, but rarely genuinely novel at the implementation level.)
I also tend not to think too much in the “agents will achieve recursive self-improvement, and all apps will just be a front end to a chat interface” direction. Not necessarily because I don’t believe it’ll happen, but because it’s the most boring possible outcome. If AI can just improvise whatever application a given user needs, then we’re truly done and can all pack it up, except for a handful of harness engineers and infra guys to keep the fires burning under the One True Application.
So where does that leave me?
On a personal level, AI is a godsend. It automates exactly the part of the work I disliked — the syntax and the hunting for the right API docs and the stray-semicolon-that’s-breaking-the-build part — and lets me focus on the parts I care about, the feature flow and the higher-level architecture, without being distracted by the minutiae. I can think about “which tests does this feature need” without having to think about “what are the syntax requirements of this testing environment.”
(I suspect a lot of the developers who are ambivalent about AI actually enjoy coding for its own sake, and that’s the part that AI is taking away from them.)
Between the corporate job and the projects I’ve been Making For Fun over the past few weeks, I’ve worked with it enough to feel I could build and support pretty much any application, up to a certain scale. I built a SaaS app just to see if I could. I could. If I have a personal need, I can build it instead of buying it. Is it me building it, or is it the AI building it? Don’t care. Inconsequential question. Between the two of us we can support it.
(The only real concern is the dependence on the hosted paid service that is the coding agent; I can only hope that the availability of compute and improvements in open models will let us all self-host someday at a Good Enough™ level.)
When I’m coding, AI is an infinitely patient teacher. If I’m confused about something, I can just ask; it has the same context I do, and all the docs available. It’s a great “rubber duck”: I can ask it to help me weigh options and get a helpful answer. It’ll point out ramifications I didn’t think of, and keep track of contradictions. (This has happened to me frequently enough that I find the “it’s just repeating its input” criticism a bit silly.) It’ll follow existing code standards, or propose sane new ones when they’re justified.
So, yeah, on a personal level, this is pretty fucking great.
Societally, it’s a fucking disaster. The techies are on their way to writing a huge percentage of themselves out of existence. The layoffs have already begun; this is just a question of degree. Then the remainder will gradually build tools and MCP servers to automate everyone else out of work too. If your job isn’t physical or in person, it’s at risk of at the least major change. [1] In the worst plausible future, the working pool becomes a game of musical chairs to see who’ll be the last one holding on, twiddling the AI’s knobs; and our economic system is not built for the needs of the majority of us who’ll be left on the outside.
That’s kinda hyperbole but kinda not. In accepting that the AI can write and glue together functions as competently as I can, I’ve had to accept that yet another set of skills I’ve spent decades of my life depending on are now redundant. (It’s just luck that in my case those particular skills weren’t the ones that keep me excited about the workday.)
That’s a hard thought to wrap up in a tidy conclusion.
And then: robots.
↩