AI is getting better at writing code every month. I asked people in tech, on Reddit and in my own network, what comes next.
Say you've just been hired by two clients in the same week.
The first is a yoga studio. The second is a clinic. Both want an app.
You have an AI assistant, so you describe the studio's app: class booking, passes, a calendar. Twenty minutes later it works. Then the clinic's: a chart that shows a patient's lab results over time. Another twenty minutes, and that works too.
You feel like a genius. (You're not, the AI is. But let's enjoy the moment.)
Six months later, the problems start. Nobody reports a bug, because there isn't one.
The studio. Your app stores a pass as a number: "classes left: 7." It looks right. Then a customer asks for a refund on a pass with four classes left. Another pass expires after a long illness freeze. A third was bought at a discount, so each class is worth less than the others. The owner needs to know what each remaining class is worth, when it expires and what it was bought for. A counter can't answer any of that, and it threw the history away. A studio owner knows a pass is really money with an expiry date, so it needs a ledger, not a number.
The clinic. Your chart plots each result against its date. Looks perfect. But a lab result has more than one time: when the sample was taken, when the result was ready, and when it was reported. Your chart used the wrong one, so a patient's glucose appears to have improved days after it actually did. Nobody asks which date, because everyone assumes there's just one. Someone who works in health data knows there are several, and knows which one matters.
Your AI never knew. Nobody told it. Nobody told you either.
Neither app has a bug. The code is fine. What's missing is knowing the work.
That's the question I keep coming back to. AI is getting better at writing code every month. So what should the rest of us learn next? And can we even keep up?
Why I Asked
I work in software, and lately everyone around me is nervous about AI. Jobs, juniors, founders wondering who to hire, the whole thing.
So instead of guessing, I asked. I posted three questions on Reddit, read another thread where people were discussing the same worry, and messaged a few people I know.
The questions were simple:
How fast is AI moving, and what will it do to tech and IT, including health tech?
Will it hurt the job market, and should people pivot?
Will domain knowledge matter more than tech knowledge in 3 to 5 years?
NOTE: This is not a survey or a study. It's a few dozen opinions from the internet and a couple of colleagues. Treat it as a conversation, not a statistic.
Computers Used to Be People
One comment stuck with me more than the rest.
A commenter pointed out that "computer" used to be a job title. People sat and did calculations all day. Then machines arrived and the work changed. People didn't vanish, but they went from doing the calculations to working around the machines.
The real history is more complicated than a Reddit comment, but the idea is a good one.
Could AI be the next computer? Maybe. Nobody knows. But it's a calmer way to look at it than "we're all doomed."
What People Actually Said
The answers fell roughly into three groups.
The "it's a tool" group. A backend developer with eight years of experience put it plainly:
"AI is a tool that makes me faster, not a replacement."
Another commenter, who said they work at a large tech company, claimed they can now do the work of five people, and summed it up as "code is cheap thinking is hard."
The "I'm worried" group. One person said their role changed from "development to developer+QA" in a year, and the QA team was laid off. Another asked a question that nobody really answered:
"What to upskill in? AI already performs the new skill I learn better than I can."
The "nobody knows" group, and my favourite honest answer of the lot:
"I don't know, neither do you, nor anyone else."
The Question Nobody Wanted to Answer
My third question was the one I cared about: will domain knowledge matter more than tech knowledge in three to five years?
Almost nobody answered it directly, which is probably because it's a hard question. A few came close, from different angles.
One commenter said you won't need to write code, but you will need to understand systems:
"you don't need to know how to code, but you do need to know how to architect"
A developer said the AI writes better code than he does, but he still has a role:
"I still need to be there because I have tastes and preferences."
A student suggested building in a field you already understand:
"get into any domain that you already have deep knowledge of"
A professional I know who works with health data said the code itself is getting cheaper, but what stays scarce is knowing the problem, which in health data means understanding how clinical and operational work really happens. They added that you still need enough tech to judge what the AI hands you.
And someone I work with, outside the tech world, wondered whether anyone can keep up with this pace at all. Their advice was the shortest I got: just have a skill around it.
So, What Should the Rest of Us Learn Next?
Back to the question in the title. After reading all these views, here's the answer I keep arriving at.
Code is getting cheaper. Almost nobody in these threads argued that it's getting harder to produce, not even the calm ones. What they kept pointing to was everything around the code: the problem, the flow, the architecture. Different people, different words, the same direction: understanding that goes beyond the syntax.
Remember our two apps. Domain knowledge isn't remembering a detail. It's knowing what the generic tool gets structurally wrong. A studio owner knows a pass is money with an expiry date. Someone in health data knows a lab result has more than one time, and which one to trust. The AI can build either one properly once someone tells it what's missing. The hard part is knowing what to tell it.
There's a twist that's easy to miss. AI has let people with limited technical knowledge explore the code side of things. The studio owner from our story could now build a rough booking tool herself. It works the other way too. AI is a helping hand that can look after part of the coding, which means people in tech now have room to walk into the domains.
So here's my answer:
Pick a niche you already know more than the basics about. Fall in love with its problems and its flow. Ask the questions nobody asks, learn how it works inside and what good architecture looks like for that kind of system. Then use AI tools to build more complete software, with your domain knowledge steering and your technical foundations checking the result.
If you're a founder, this matters just as much. It decides who you hire and what you build.
Can We Keep Up?
If keeping up means trying to match AI at writing code, I don't think that's a race worth running. It's getting faster, and the gap will probably keep growing.
If it means knowing your field deeply enough to tell the right thing from a fast thing, then yes. That's something people can still build, one question at a time.
The people who have both will be the ones building the right thing, not just a fast thing.
That's my read, anyway. It's not a prediction, just where a few dozen strangers, a few colleagues and two invented clients seem to point.
What would you add? Tell me in the comments. I promise to read them.
And if you made it to the end, you now know why a pass isn't just a number 😉
I build custom software and automation for clinics and small businesses, usually starting by learning how the work really runs. If that sounds useful, say hello at [www.vishal@observingowl.com].
