There’s an interview question I ask every developer candidate: “Tell me about a time you were wrong about a technical decision.” I’ve written before that the honest answers are how you spot a growth mindset — people who can describe what they missed, what they learned, and what changed.
Fair is fair. Today I’m answering my own question.
In February 2023, I published What I’d Tell Someone Starting a Software Development Career Today. I wrote it about ten weeks after ChatGPT launched, when AI coding tools were — as I put it at the time — autocomplete on steroids. I’ve patched that post since; if you read it today you’ll find a revised AI section that already admits how much changed. But patch notes aren’t the same as a reckoning. Three and a half years is long enough to grade the whole thing honestly: what aged well, what needed the quiet edits, and what I’d say differently starting from scratch in 2026.
The parts that aged better than I expected
The throwaway paragraph turned out to be the thesis. In the section about picking a first language, I wrote that the specific language matters less than the habits you build learning it: Can you read error messages without panicking? Can you break a problem into smaller problems? Can you search for answers effectively? I meant it as advice for choosing between JavaScript and Python. Reading it now, I accidentally described the job of working with AI. Decomposing a problem, evaluating an answer you didn’t write, staying calm in front of output you don’t yet understand — that’s the daily loop of AI-assisted development. I’d promote that paragraph from aside to headline.
The day-in-the-life section was more prescient than I knew. I described a typical day to counter the “designing, building, testing software” boilerplate: reviewing a teammate’s pull request, spending two hours reproducing a customer issue for a twenty-minute fix, pairing with a junior, communicating progress. Then I pointed out how much of that day was reading, communicating, and investigating rather than writing new code. In 2023 that ratio surprised newcomers. In 2026 it has shifted even further — the pull requests you review now often come from an agent, and the investigating and judging is more of the job than ever. I described the editor’s job before I had the word for it.
The skills list needs no edits. Learning in public, communication, persistence through frustration. In 2023 I called these the skills that matter early on. Today I’d call them the skills that matter, full stop. When implementation is increasingly delegated, what remains is asking good questions, explaining your reasoning, and staying curious when something breaks. Those were accelerants then. They’re the core now.
The parts that held, with caveats
The degree advice stands — the “why” got stronger. I told my own story: years as a working developer before finishing my B.S. in 2017, recruiters who filtered me out, the gaps that structured education eventually filled. All still true, and I’d still say start without the degree if you have to and finish it if you can. What’s changed is the value of what a good program actually teaches. Data structures, algorithms, systems design — I framed those as foundations that are “hard to pick up piecemeal.” Now they’re the substance of the judgment that AI can’t supply. The credential opens doors; the fundamentals are what let you tell when a generated solution is wrong. The second thing matters more than it did.
“Pick a language and go deep” survives, but the reason changed. In 2023, depth in one language was the practical path to employability. In 2026, language-specific fluency is the cheapest it has ever been — any capable model writes idiomatic code in any mainstream stack. Going deep in one language is still right, but not because you’ll be paid for the syntax. It’s because depth in something is how you develop the instincts to evaluate everything else. You learn what correct feels like somewhere specific before you can judge it anywhere.
The parts I got wrong
The pace, obviously. The original AI section answered “is it still worth becoming a developer?” from a world where the tools autocompleted lines. I’ve already revised it once as agents started drafting features and working multi-step tasks with limited supervision — and I’ll defend the revision, especially the hard truth it names: AI is raising the bar for entry-level work because the tasks juniors cut their teeth on are the tasks AI handles first. But I want to grade the pattern, not just the paragraph. I underestimated the pace once, corrected, and the correction is already aging. The lesson I’m taking is structural: any career advice indexed to what the tools can’t do yet has a shelf life measured in months. Advice has to be indexed to what doesn’t move — judgment, ownership, fundamentals — or it needs a revision schedule.
The “Where to Start” list assumed a pipeline that has narrowed. Finish a curriculum, put projects on GitHub, fix typos in open source, apply before you feel ready. None of that is wrong, but in 2023 it quietly assumed that a portfolio of completed-curriculum projects was a differentiator and that entry-level funnels were wide. Both assumptions weakened. A folder of tutorial builds proves less when AI can generate tutorial builds; documentation-typo contributions prove less when the signal employers need is judgment, not activity. If I rewrote that list today, step two would change from “build something you care about” to “ship something whole”: a real thing with real users — even five of them — that you designed, built, deployed, broke, and fixed. End-to-end ownership is the credential that survived, because it’s the one thing you can’t generate.
And “apply before you feel ready”? Still true. You’ll just need more applications than the 2023 version of me implied, and the readiness that matters has changed shape: less “can you code” and more “can you own.”
What I’d add, starting from scratch in 2026
The revised post already tells newcomers to use AI to learn faster rather than to avoid learning — have it explain code line by line, ask why an approach works, then close it and rebuild the thing yourself. I stand by every word; that’s pilots practicing hand-flying, and it’s how judgment gets built now.
What I’d add is the view from the other side of the hiring table. The industry spent decades treating code as the scarce output and judgment as something you picked up along the way. That’s inverting. The scarce thing now — the thing I can’t hire enough of — is people who can decide what to build, verify it was built right, and take responsibility for it in production. That’s a harder job to start than the one I described in 2023. It’s also, honestly, a better one: you skip years of typing boilerplate and go straight to the part of the work that was always the point.
I was wrong about the pace once, so hold that prediction as loosely as I should have held the first one.
This post revisits What I’d Tell Someone Starting a Software Development Career Today (February 2023, since revised).
