"Learn AI" is not a career plan. It's a tool you pick up on the way to one.

Developers starting out keep asking some version of the same question: is it still worth learning to code properly, now that a model writes code? Yes. But the question is aimed at the wrong thing.

I didn't plan a career in software. In 2014 I was a high-school hobby photographer who needed a website for his pictures. Every step since has taught me the same lesson: the skill gets you into the room, and something else keeps you there.

What got cheap, and what didn't

Before

Cheaper for everyone

  • Basic coding
  • A first draft of almost anything
  • Boilerplate, tests and documentation drafts
  • Output that looks like it works at first glance

After

Still scarce

  • Expertise: knowing what will break, and where
  • Judgment: knowing what not to build
  • Responsibility: owning the outcome when it goes wrong
  • A working result: used by real people, not just demoed

AI makes producing code easier for everyone, which makes basic coding services more competitive and cheaper. If your plan is to sell what a model now produces in seconds, the model is your competition, and it never gets tired.

So the opportunity isn't that you can use AI. Everyone can. It's connecting AI to real domain knowledge, solving the actual problem, owning the outcome and shipping something that works.

Knowledge gets you in. Reliability keeps you in.

~0

year I've worked as an employee, in my whole career

and even that was contract work

Freelancing for local businesses while I studied software engineering, a studio that grew past ten people in 2019, then back to the craft in 2021 with international software companies. Through all of it, my work depended on clients choosing me again next month, and that teaches you fast what actually keeps you in work.

  1. 1ReliabilityWhat you promise, you deliver. Over the long run that's worth more than any impressive introduction, and it's why clients come back.
  2. 2CommunicationYou understand what the client actually needs, and you ask when you don't. A project built perfectly on a misunderstanding is still a failed project.
  3. 3Fast learningWhat was an edge last year is the baseline this year, AI tools very much included. Learning fast is a habit you build, not a talent you're born with.
  4. 4ResponsibilityWhen something slips, you're the one who says it first, before anyone has to ask.

AI multiplies these four. It doesn't replace any of them. A reliable engineer with AI becomes far more valuable; an unreliable one just produces unreliable work faster, and in larger quantities.

Communication is a professional skill

It's easy to believe that if you're good enough technically, everything else sorts itself out. It doesn't. Clients rarely care about the technology; they care whether you understood what they're trying to achieve, whether you can deliver it, and whether they'll hear about a problem from you before they find it themselves.

  • Asking before you build, not after
  • Explaining the trade-offs in the client's words, not yours
  • Flagging a slip the day you see it
  • Saying out loud when the original idea won't solve the problem
  • Setting expectations you can actually keep, and then keeping them

Good communication isn't selling yourself well. It's the other person knowing exactly what to expect from you.

Picture two developers with the same skills: one builds exactly what was asked, the other notices that a simpler, cheaper solution would solve the problem and says so. Over time, the second one wins, because clients remember who saved them money. AI makes that gap wider, because building exactly what was asked is the part that got cheap.

Proof over polish

AI makes impressive-looking material cheap: the slick portfolio, the confident post, the demo no real user has ever touched. That's exactly why proof is worth more every month. You don't need ten thousand followers. You need this:

  1. 1Three good projects. Finished, and used by someone other than you. Three real ones beat thirty tutorials.
  2. 2Clearly explained. The problem, what you did, what changed because of it.
  3. 3Shown to people whose problem you can solve. One good recommendation or one past client's trust is worth more than reach. Visibility isn't performance.

If I were a student today

Where I'd put my time
  1. 1A skill that solves a real problem
  2. 2AI in my own daily work
  3. 3Talking to people, showing work

Programming, design, sales, automation or a niche specialty: not the trendiest one, the one you can create value with. Not everyone needs to become an AI expert, but everyone needs to use it in their own work. And don't stop at courses: build something, show it to someone, ask for honest feedback, fix it, and repeat until the loop feels normal.

"I'll start when I'm ready" is the most expensive sentence in a career. Nobody is ever ready. Start small, with one project or one first client, while the stakes are still low and a mistake costs you a weekend, not a reputation.

One of the principles I keep for myself is three words long: imperfect > nothing. Work that ships imperfect beats perfect work that never ships, and the imperfect version is the one that teaches you what to fix next.

AI will speed up whatever you're building. It won't decide what that is. That part is still your job.

Weekly-ishEngineering OS, in your inboxJoin Engineering OS