In February 2026 the engineer and writer Marina Wyss published a short Medium piece, “Should You Still Learn to Code in 2026” — an honest take on the profession two years after LLM coding assistants moved permanently into the IDE. I read it a few times, agreed with some of it, pushed back on the rest, and decided to write my own walk-through: where she is right, where she over-simplifies, and what I would add.
What the original argues
Wyss states what is now the loudest claim in the industry: writing code is no longer the central task. The assistant generates most of the syntax; the developer investigates, verifies, explains, negotiates. In her experience the entry door has narrowed — open positions have roughly halved, junior roles in particular. Yet the motivation to learn is not invalidated: as she puts it in one of her stickier lines, “AI won’t wake up. You will.”
Where I agree
Her core observation feels right. The value of an engineer in 2026 has shifted from “writes code” to “understands what should happen.” When the assistant can spit out any function in seconds, the winner is the one who can phrase the right question, read a diff, and notice the part that doesn’t fit. This isn’t a new skill — it’s old-school programming — except its share of the job has grown.
“AI won’t wake up. You will.”
That sentence catches what many writers try to dress in five paragraphs. The discipline of attention, and the habit of reading other people’s code, are now the muscles that matter.
Where I’d add caveats
Wyss sometimes treats “learning to code” as one activity. In practice the phrase carries at least three meanings, and 2026 looks different for each:
- Learning to land a first job. Here she is right: the junior market has tightened, and finishing a bootcamp no longer opens the door on its own. What works is visible projects with a clear interest profile.
- Learning to understand your own product. Product managers, analysts and designers who can read SQL through to the end and fix a query win bigger than they did three years ago. AI doesn’t replace them — it amplifies the ones who already had technical literacy.
- Learning for the task itself. Hobby programming hasn’t gone anywhere, and the LLM assistant makes it more accessible: the barrier of “syntax in the way of the idea” is gone. That’s a new, quiet mode of learning — and her piece barely mentions it.
What the article leaves out
There’s an unspoken assumption that the 2026 developer is the same person as the 2020 developer, just plus an assistant. The career path is changing too: the vertical ladder of junior → mid → senior is flattening. Often it’s more useful to move sideways — into an adjacent domain, infrastructure, data, security — and carry the code-reading skill with you. AI makes narrow specialisation riskier: anything a model does well is now done for free.
The other thing I didn’t find in the original: a candid discussion of tool dependency. When the assistant writes 80% of the code, you slowly lose the reflex of “open the docs, read the page.” At first it feels like speed; eventually it feels like a musician losing the ear. It’s worth deliberately working old-school from time to time — by hand, no prompts.
The short version
Wyss asks the right question and gives an honest, practical answer for anyone considering the profession. I’d add that “learning to code” is not one story but at least three, and for two of them 2026 is a fine year to begin. The trap is confusing “I’m learning the syntax” with “I’m learning to think like an engineer.”
The full original is on her Medium. If the 2026 conversation interests you, read hers before mine — it is shorter and more decisive.