Krouse argues that coding education was never fundamentally about typing syntax — it was about learning to decompose problems, model systems, and reason about state and failure. LLMs are eating the keyboard-mechanical part, but the reasoning skills that separate working software from plausible-looking software are exactly what coding literacy builds.
Krouse frames coding literacy the way earlier generations framed reading: not everyone needs to be a novelist, but a society where nobody can read gets lied to easily. In an AI-mediated world, the ability to inspect and reason about code becomes a form of protection against being deceived by systems you can't understand.
The editorial argues that CEO takes dismissing coding education make a specific category error: LLMs are good at producing code but bad at comprehending it. Asking Claude to write a component works; asking it to explain why a colleague's component leaks memory under concurrent renders produces confident guesses that fall apart on inspection.
The HN discussion largely agreed with Krouse but repeatedly qualified the endorsement — the curriculum, tools, and expected outputs of a coding education have shifted enough that agreeing 'coding still matters' doesn't settle what should actually be taught. The consensus was on the value of the reasoning skill, not on the specific pedagogy that produces it.
Steve Krouse — the founder of Val Town and a long-time end-user-programming thinker — published a short essay titled *Learning to code is still worthwhile* that hit 139 points on Hacker News. The piece is a direct response to a growing genre of takes from CEOs and pundits telling kids (and career-switchers) not to bother learning to program because LLMs will handle it.
Krouse's argument, compressed: learning to code was never really about typing syntax. It was about learning to decompose problems, hold a system in your head, and reason about state, failure, and change. The keyboard part is what LLMs are eating. The reasoning part is what still separates people who ship working software from people who ship plausible-looking software.
He frames coding literacy the way earlier generations framed reading: not everyone needs to be a novelist, but a society where nobody can read is a society that gets lied to easily. The HN thread — which stayed near the top of the front page for most of a day — largely agreed, though with the usual caveats about whether "learning to code" in 2026 means the same thing it meant in 2016.
The "don't learn to code" discourse has been running for about eighteen months now, kicked off by Jensen Huang's 2024 comment that coding is dead and accelerated by every subsequent model release. It's mostly wrong, and it's wrong in a specific way that matters to anyone hiring or being hired.
The error is treating code production and code comprehension as the same skill. LLMs are genuinely good at the first and mostly bad at the second. Ask Claude or GPT to write a React component and you'll get something that runs. Ask it to explain why the component your colleague wrote last quarter is leaking memory under a specific concurrent-render pattern, and you get confident guesses that fall apart on inspection. The bottleneck in software has never been typing speed; it's been the mental model of the system, and no model to date has convincingly shipped one of those.
Krouse's essay lands harder because he's not making the reactionary version of this argument. He runs Val Town, a platform whose entire pitch is *lowering* the barrier to writing code — much of it now AI-assisted. He's not defending gatekeeping. He's saying that even in a world where the first draft comes from a model, the person who can read the draft, spot the bug, and steer the next iteration is doing something the model can't do for itself. That person needs to have learned to code the old-fashioned way at some point. There's no shortcut to taste.
The HN comments surface the honest counterpoint: the on-ramp is genuinely different now. A beginner in 2026 who leans on Cursor or Copilot from day one may never build the debugging muscles that people who learned in 2010 built by necessity. Several commenters — including some teaching intro CS — reported students who can prompt their way to a working assignment but can't explain what a for-loop does when you take the autocomplete away. That's a real problem, and it's not solved by telling those students they were right not to bother learning.
What this points to is a skills bifurcation the industry hasn't fully priced in yet. On one side: people who can ship features by orchestrating models, without a deep model of what the code does. They're productive on greenfield work and near-useless on incident calls. On the other: people who understand systems well enough to catch the model when it hallucinates a package that doesn't exist or a security assumption that isn't true. The second group's market value has, if anything, gone *up* since GPT-4, because they're now the only people who can safely deploy the first group's output.
If you're hiring: stop screening on ability to write code from scratch on a whiteboard — the LLM will do that on the job anyway. Screen on ability to read unfamiliar code and reason about what it does wrong. Give a candidate a 200-line file with two subtle bugs and ask them to find them. The ones who can are worth 3x the ones who can only prompt.
If you're on a team: the review process is now the choke point. AI-generated PRs are cheap to produce and expensive to verify, and the verification burden falls on humans who understand the code deeply. Teams that let LLM output through review without someone who actually understands the change are shipping tech debt at machine speed. Budget for it — either by slowing the throughput or by explicitly rotating strong reviewers onto AI-heavy work.
If you're early in your career and reading the "don't learn to code" takes: ignore them, but adjust the curriculum. Build one real thing without AI assistance, end to end, so you know what the layers actually do. Then use AI liberally after that. The order matters. People who start with the abstraction never learn what it's abstracting.
If you're a senior dev worried about being commoditized: you're probably worried about the wrong thing. The commoditizable skill is *writing* code, and it was already halfway commoditized by Stack Overflow. The non-commoditizable skill is knowing which code should exist, which shouldn't, and why the model's suggestion is subtly wrong. That's what you've been training on for a decade. It just got more valuable, not less.
The next twelve months will settle whether the "learn to code" question is still even the right frame. Krouse's implicit bet — and it's a defensible one — is that the durable literacy is *computational thinking*, not any specific language or tool. If he's right, the winners in a coding-agent world will be the people who can specify problems precisely, evaluate solutions honestly, and hold a system model in their head across sessions. Those are all skills you build by writing code. Painfully, with bugs, without autocomplete. The essay is worth the ten minutes it takes to read, especially if you have a junior on your team who's been told they picked the wrong career.
> Code is a beautiful form of creative expression, as rich as literature or musicI think this is overstating it and makes me wonder how familiar the author is with literature and music. Most programming is closer to plumbing. We come in, gripe about the guy who did the prior job, and solve a puzz
As a professional programmer entering the final third of an enjoyable career, I would now place "learning to code" in the same category as "making a living as a poet." As in, it's truly enjoyable art and some people appreciate it, but you'd better plan for a day job.Sen
This “don’t need to learn code” vibe will bring a big quality and good pay wave programming. It was getting more obnoxious because of more and more mba guys getting into programming because there is money in it.Hopefully this will act as an effective filter against money people so they can keep goin
> Code is a beautiful form of creative expression, as rich as literature or musicSomething I'm trying to do right now is to build something and avoid using LLMs to write any code. I still use it to consult. I'm writing a Dota2 tournament match aggregator in Elixir that takes tournament
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
I think that, even among those who have had the ability to code since before LLMs, the skill of coding has been in atrophy, and will continue to be so over the next 10 or 20 years. Learning to code, over the same period of time, will not be worthwhile. At the end of that process, cracks in the syste