AI is doing my old job now — and honestly I have mixed feelings about it

Last month I asked an AI coding assistant to migrate a legacy authentication module I used to maintain — the kind of job that would have eaten two of my days as a senior engineer, most of it spent re-reading old logic to make sure I understood every edge case before touching it. It had a working draft in about four minutes. Not a perfect draft. It missed a rate-limiting edge case I had to catch myself, and the error handling needed real revision. But four minutes, for what used to be two days, is not a rounding error. It's a different category of thing.

I want to write about what that actually felt like, because the emotional experience of watching this happen in real time is more interesting, and more honest, than any prediction I could make about where it all ends up. The prediction business around AI and engineering jobs is full of confident people who don't have to live with being wrong. I'd rather describe what it's actually like to sit with a tool that does part of your old job well, and feel several contradictory things about it at once.

The task that started this

I left a senior engineering role a little over a year ago, for reasons that had nothing to do with AI — burnout, mostly, the ordinary kind this whole site is about. I still do occasional contract work in the same domain, which is how I ended up back inside that old authentication module, watching a tool do in minutes what had been a meaningful chunk of my professional identity for a decade.

The first reaction was purely practical relief. The work was tedious, the kind of task that's necessary but not where the interesting thinking happens, and having it accelerated felt like exactly what better tools are supposed to do — the same relief I'd have felt about a good linter or a solid test suite, just a larger version of it. If I'd still been employed full-time and under deadline pressure, that four minutes would have been an unambiguous gift, and I want to be honest that the gift was real before I get into the more complicated part.

What came after the relief

The complicated part arrived about an hour later, once the immediate task was done and I had time to think about what I'd just watched. The thought wasn't "I'm about to lose my job" — I wasn't in a job to lose, and the work I do now doesn't compete directly with what the tool did. The thought was quieter and, in some ways, harder to sit with: a decade of accumulated skill at exactly this kind of task had just been made meaningfully less scarce, and scarcity was a real part of what that skill used to be worth, both economically and to my own sense of being good at something specific.

This is the part I think gets flattened in most discussion of AI and engineering work, in both directions. It's not simply "my job is safe" or "my job is gone." It's a specific, uncomfortable recalibration of how much of what I spent years building is now something a tool can approximate well enough that the approximation is genuinely useful. Not identical to what I could do. Good enough that the gap matters less than it used to for a meaningful share of the work.

The redundancy question underneath the relief

Here's the honest version of the anxious question, stated plainly rather than dressed up as a hot take: if a tool can do the mechanical middle of engineering work — the migration, the boilerplate, a first draft of the tricky function — at a fraction of the time, what happens to the value of having spent a decade getting good at exactly that? I don't think there's a single honest answer to this yet, and I'm suspicious of anyone who tells you they have one, in either direction.

What I can say, from watching it happen at close range rather than reading about it from a distance, is that the tool was very good at the part of the work that had a clear, bounded shape and a lot of precedent to draw from, and considerably less reliable at the part that required knowing which edge case actually mattered in this specific system, with this specific history of small compromises nobody had documented. The rate-limiting bug it missed wasn't a hard bug. It was a bug you'd only catch if you'd been burned by a similar one before, in a codebase you actually understood the shape of. That kind of catch is exactly the thing a decade of doing the work produces, and it's also exactly the kind of thing that's hard to notice you're contributing until a tool shows up that's good at everything except that.

The evidence so far suggests this pattern holds more broadly than my one anecdote: these tools are genuinely strong at generating plausible, well-formed code quickly, and meaningfully weaker at knowing when the plausible answer is subtly wrong for reasons specific to the system it's operating in. Whether that gap closes, narrows, or stays roughly where it is, is a real open question, and I don't think the honest answer is currently knowable with any confidence. Anyone stating it as settled — in either direction — is telling you more about their own priors than about the technology.

The identity question, which is harder than the economic one

The financial question — will there be as much demand for this specific skill set at this specific price — is genuinely uncertain and worth taking seriously without resolving it into false confidence. But it turned out not to be the question that sat with me the longest. The harder one was about identity, and it's connected to something I've written about elsewhere on this site: how much of a sense of self gets built quietly on top of "I am good at a specific, scarce thing," without you noticing the foundation until something shifts underneath it.

Watching a tool do a compressed version of my old job well didn't just raise a question about future income. It raised a question about what all those years of accumulated, hard-won competence were actually for, if a meaningful chunk of the visible output can now be approximated quickly by something that didn't have to live through any of the mistakes that taught me how to do it. I don't have a tidy resolution to that. What I have is a suspicion, not yet fully tested, that the years weren't wasted — that the judgment behind the rate-limiting catch is still mine, still earned, and still not something the tool has, even if the typing part increasingly is. But I'm aware that's the story I'd want to be true, and I'm trying to hold it as a hypothesis rather than a conclusion.

What I've landed on, for now

I don't think the honest position is "don't worry, AI is just a tool" — that dismisses something real happening to a real category of work, and I think readers of this site, most of whom have spent years in tech, will correctly distrust that reassurance as thin. I also don't think the honest position is "your career is being erased" — that's a clean, alarming story that fits a headline better than it fits what I actually watched happen at my old keyboard, which was more specific and more mixed than either extreme.

What I actually believe, provisionally, watching this from close range rather than from a headline: the mechanical middle of engineering work is getting genuinely, rapidly easier to approximate, and that's not a false alarm. At the same time, the specific judgment that comes from having been wrong many times, in ways a fast tool hasn't yet been wrong, still seems to matter — for now, in the systems I've watched this play out in, at the current state of these tools. I'm holding both of those as true simultaneously, and resisting the pull to resolve the discomfort by picking the version of the story that lets me stop thinking about it.

If you're inside a tech career right now and watching a similar version of this happen to your own work, I don't think there's a script that resolves the discomfort quickly, and I'd be wary of anyone offering you one. What I can say is that the discomfort itself — the mixed feelings, the identity questions, the not-knowing — is a reasonable response to a genuinely uncertain situation, not a sign you're failing to understand something that's actually simple.

The piece on what happens to your tech skills when you step away covers some of the same territory from a different angle — including the specific point about AI tooling and judgment that this piece builds on. The piece on what happens to your personality when you stop being "the smart one in the room" is the closest companion to the identity question raised here. And if you're trying to work out what you'd actually want to do next, independent of what AI does or doesn't do to your current role, the five-question framework for big career decisions is a useful place to start.