Bet Either Way: AI Will Win and Almost Every Programming Job Will Vanish—or There Will Be a Third AI Winter
We keep hearing two confident predictions about artificial intelligence.
The first says AI will keep improving until it can do nearly all programming work, making most programming jobs disappear. The second says today’s excitement is another bubble: progress will disappoint, investment will cool, and we will enter a third AI winter.
I don’t know which future will arrive. Both are possible, and each is worth taking seriously. But the best response may not depend on choosing the right prediction. It may be to prepare for either one.
There is also a question underneath both forecasts: what does it mean to be a programmer when writing code is no longer the hardest or most valuable part of building software?
If AI Wins: Tools Become Abundant
Suppose AI continues to improve. It becomes cheaper, more capable, and better at building software from a description of what someone wants. Many tasks that now require a developer—writing routine code, adding tests, fixing common bugs, or assembling a simple application—could become much easier to automate.
That would change programming work dramatically. It still would not automatically mean that every programming job disappears. Software work includes understanding what people need, deciding what should be built, connecting systems, protecting data, checking whether a result is reliable, and maintaining it over time. AI can help with these tasks, but a working demo is not the same as dependable software.
Still, if AI makes building software much easier, the number of people who can make their own tools will grow. Someone with enough knowledge, time, or money may be able to describe a need and create a useful application for themselves. A small business might build an internal tool without hiring a full development team. An individual might make a personal workflow that once would have seemed too expensive to commission.
This leads to a blunt warning: making tools alone may not make you rich.
If everyone can create a tool, the mere fact that you made one will no longer set you apart. The tool may be useful to you and still have little commercial value. For a product to succeed, it has to solve a problem that matters to other people, reach them, earn their trust, and keep working as their needs change.
AI may lower the cost of making software without lowering the difficulty of finding customers. The scarce skill could shift from writing code to recognizing a worthwhile problem, understanding a particular community, delivering a trustworthy solution, or building a business around it.
Will We Pay for Personal Software?
One possible future is a market for handcrafted personal software.
Today, people can buy ready-made tools or hire someone to build custom software. If AI reduces the time and cost involved, a new kind of service might emerge: a person or company that learns how you work and builds a tool fitted closely to you. It could be software equivalent to a high-end tailor, or to a company that makes a car to a customer’s specifications.
That comparison is about the experience of customization, not a guarantee that the economics will resemble luxury fashion or Ferrari. A personal software service might be affordable and practical; it might also become a premium offering for people who value close attention, reliability, and ongoing support.
The hard part will be more than generating code. A provider would need to understand the customer’s work, keep private information safe, make the software dependable, and update it as the customer’s needs change. The value may lie in that relationship and ongoing service as much as in the software itself.
This is only one possibility. Perhaps people will prefer flexible general-purpose tools. Perhaps they will make their own applications with AI. Or perhaps personal software will become a normal service, no more remarkable than hiring a professional to tailor something important to your needs. We do not know yet.
If This Is Another AI Winter: Use the Time to Learn
The other possibility is that today’s expectations outrun what the technology can reliably deliver. AI could remain useful while failing to meet the grandest promises. Costs, limitations, disappointing results, or a slowdown in investment could cool the current excitement.
An AI winter would not mean that all progress stops or that every AI product disappears. It would mean that enthusiasm and funding retreat, and that some of the predictions made during the boom prove too optimistic. That could happen even if the tools continue to help with real work.
If that happens, the quieter period could be a good time to learn programming.
But I don’t think learning should mean memorizing syntax for its own sake. AI can already explain concepts, suggest exercises, review code, and help plan projects. A learner can ask it to make a sequence of small challenges, starting with a simple program and gradually adding features. That can support project-based learning: build something, encounter a problem, learn the concept needed to solve it, and improve the project.
This could help people avoid tutorial hell—the cycle of watching courses and following examples without becoming able to make something independently. A project gives learning a purpose. It also makes progress visible: the thing works, breaks, changes, and eventually becomes more useful.
But AI guidance is not automatically a good education. It can give incorrect explanations, write code the learner cannot understand, or remove the useful struggle of solving a problem. A learner should ask for hints before full solutions, explain the code back in their own words, test what they build, and deliberately try to fix problems without copying the first answer.
In that sense, AI does not make programming books useless. It changes what a book is for. A book can still provide a carefully edited explanation, a durable reference, or a coherent path through a subject. But for many learners, an interactive tutor that adapts a project to their interests may be a more engaging place to begin.
Learn to Understand, Not Just to Produce
If AI can generate code, learning programming still matters—but the goal may change.
Understanding code helps you tell whether an answer is plausible, find the cause of a bug, protect sensitive data, and decide whether a program is doing what you intended. It also gives you a way to make changes when the first generated version is almost right but not quite.
The learner’s advantage may come from combining technical understanding with judgment: asking good questions, breaking a large problem into parts, evaluating the result, and knowing when to keep digging. AI can speed up the work, but speed is useful only when you can recognize a good result.
The Bet May Not Be the Point
The two forecasts—AI changes programming beyond recognition, or the current wave cools into another winter—are useful scenarios. Neither is a settled fact.
Current evidence points in both directions. AI investment, adoption, and model capabilities have grown quickly. Stanford’s 2025 AI Index reported that 78% of surveyed organizations used AI in 2024, up from 55% in 2023, and documented substantial gains on several benchmarks. At the same time, the U.S. Bureau of Labor Statistics projected growth in software developer employment from 2023 to 2033, while noting that AI could both augment programming work and create demand for people who build and maintain AI systems. Those figures describe different things—technology adoption and a U.S. employment forecast—and neither tells us exactly what will happen next.
Early labor-market evidence is also mixed. A Federal Reserve analysis found that coder employment growth had slowed sharply since the introduction of ChatGPT, but that coder employment had continued to grow. That is a meaningful warning about possible pressure on the occupation; it is not evidence that programming jobs have already vanished.
So I would be careful with the claim that almost every programming job will disappear. It is a forecast, not an established fact. We can see that AI is changing parts of the work, but the scale and timing of its effect on whole occupations remain uncertain.
Maybe AI will make software so easy to produce that many existing jobs disappear. Maybe expectations will cool, and programmers who learned how to build and reason about software will find themselves well prepared. Maybe the future will be a mixture: fewer people doing routine coding, more people using AI to build software, and new kinds of work appearing around the technology.
For now, I see value in preparing for both possibilities. If AI keeps advancing, learn to use it while building judgment that goes beyond generating code. If progress stalls, keep learning the fundamentals and make real projects. Either way, the useful question is not only “Will AI win?” It is also “What can I learn or build now that will still matter under more than one future?”
Sources