AI in software development: who will win in the job market?

A few years ago the question was: “should I use AI to write code?” Today AI in software development is part of everyday work, and that question is out of date. According to the DORA 2025 report, 90% of software development professionals use AI, with a median of two hours a day spent working with these tools. In the Stack Overflow 2026 survey, with more than 30,000 responses, 73% of people who use AI coding assistants or agents do so every day.

If almost everyone has the same tools, a different question matters more: who actually benefits from them, and why?

Our view is simple. A developer who learns to work with AI will do more, and do it better, than someone who avoids these tools because they believe hand-written “clean code” solves every problem. But simply switching on an assistant does not create that advantage. The research shows this clearly.

There are clear gains. In three field experiments at Microsoft, Accenture and a large Fortune 100 company (almost 5,000 developers, published in Management Science), people with access to an AI assistant completed 26% more tasks on average. An important point for junior developers: less experienced developers adopted these tools more readily and gained more from them.

There are also surprising slowdowns. In METR's 2025 randomised study, experienced open-source developers working in their own mature projects took 19% longer on tasks with AI. They expected a 24% speedup, and after finishing they still believed they had been 20% faster. In a follow-up study published in February 2026, the results pointed towards a speedup instead. METR is candid, however, that this data is weak evidence: more and more developers declined to take part at all because they could not imagine working without AI.

Trust is limited. In the Stack Overflow 2025 survey, 46% of developers said they did not trust the accuracy of AI output. The most common frustration, named by 66%, was solutions that are “almost right, but not quite”. 45% said debugging AI-generated code takes more time than they would like.

AI amplifies what is already there. The main conclusion of the DORA 2025 report is that AI acts as an amplifier. Good teams and good practices become even better. Weak processes and messy architecture become even more visible.

The takeaway from research on AI in software development is neither “AI always speeds you up” nor “AI is hype”. It is: the effect depends on how you work.

The most worrying data concerns the start of a career. Stanford researchers (Brynjolfsson, Chandar, Chen, “Canaries in the Coal Mine?”, August 2025) found that employment of software developers aged 22-25 in the US fell by almost 20% from its peak in late 2022. Over the same period, employment of more experienced people in the same occupation remained stable. Economists still debate how much of this is caused by AI and how much by other factors.

Poland shows a similar direction, although the data covers job postings rather than employment itself. According to the No Fluff Jobs report “IT job market 2025/2026”, the number of IT job offers grew by 44% year on year in 2025. At the same time, the share of offers for juniors fell from 8% in 2023 to 5.3% in 2025, or one posting in twenty. Almost 60% of offers target senior roles, while around 15,000 IT graduates try to enter the market every year. Among the causes, experts in the report point to company caution, the cost of onboarding new people and the rise of AI tools that have taken over some basic tasks.

Across Europe, the Linux Foundation “2026 State of Tech Talent Europe” report also describes a contraction in entry-level technical roles, alongside overall IT hiring growth that companies link to AI.

Three different markets and three different research methods, but the direction is the same: the market is growing, yet it pays less and less for simply writing basic code.

There is another side too. The same research that shows a difficult start also shows that junior developers gain the most from using AI well. A tool that takes away some simple tasks also lets you move on to harder ones sooner.

There is one important caveat. In Anthropic's study from early 2026, people learning a new library with AI assistance scored 17% lower on a comprehension quiz than those who coded by hand (50% versus 67% on average). The gap depended on how they worked, though. Those who asked AI for code and then asked why and how it works scored 65-86%. Those who simply delegated the task scored 24-39%.

That may be the most important lesson for anyone starting out: AI can speed up your learning or replace it. The choice is yours.

Discussions often split into two camps. One says AI writes the code and classic programming skills no longer matter. The other defends hand-written clean code as a foundation that must not be touched.

In our view, both sides of the debate about AI in software development confuse the tool with the goal. Clean, readable code still matters a great deal, precisely because more and more code is written with AI. Someone has to judge whether the generated solution is good, secure and consistent with the system architecture. That someone is the developer.

The difference is where the developer spends their time. Less on retyping repetitive fragments. More on understanding the problem, designing the solution, checking quality and taking responsibility for the result. A conservative who rejects AI loses time on things a machine can do faster. An enthusiast who accepts every suggestion uncritically loses control over what they ship. The advantage belongs to someone in between: a developer who leads AI rather than being led by it.

1. Describing the task precisely

AI gives back as much as you give it. A good brief includes context, constraints, the expected result and an example. It is the same skill you need to write a good ticket or specification, so it also helps when working with people.

2. Critical verification

Read, test and fix AI-generated code. “Almost right” code is more dangerous than code that obviously does not work. Tests, code review and static analysis tools are no longer a formality but a basic safeguard.

3. Thinking in architecture

AI handles a single function well. It is less good at how that function fits into the whole system, its data, permissions and future development. The DORA report shows that teams with loosely coupled architecture and fast feedback gain the most from AI. It is a skill worth learning from your very first project.

4. Learning with AI, not instead of learning

When AI gives you a solution, ask why it looks the way it does, what the alternatives are and what could go wrong. Anthropic's study shows that working this way lets you keep the pace and actually learn.

5. Understanding the business problem

Code is a means, not an end. A developer who understands why a system is being built, which process it should improve and who will use it will ask AI better questions and notice sooner when an answer misses the need. This is the skill that is hardest to automate.

1. Pick one tool and use it every day for a few weeks instead of testing them all at once.
2. Start with tasks you understand well. It is easier to judge whether AI is right.
3. Read every generated fragment as if you had to defend it in a code review.
4. Ask AI for explanations, not just solutions.
5. Check the rules for using AI in your company and project, especially when it comes to client code and data.

At our company, AI is used widely, both internally and in client projects.

Internally, we train our team and systematically build their skills in working with AI. We are also expanding our own system based on large language models, which supports how we run the company and our marketing.

For clients, AI helps us build custom software components and tools that successfully optimise and speed up work in their companies and organisations.

This experience shapes our view of AI in software development: AI does not replace a good team, but a good team with AI does more, faster and with greater control over quality.

The question is no longer whether AI in software development will change a developer's job. It already is. The real question is whether you will be among those who use AI to learn faster and build better solutions, or among those who either reject it or hand all their thinking over to it.

The research of the last two years is clear: the advantage goes to those who combine AI with solid fundamentals. Clean code, testing, architecture and understanding the problem are not losing value. On the contrary, they are what sets a good developer apart from a well-written prompt.

For a broader look at how AI is changing the job market, read our article Will AI replace employees? Facts and myths.

If you want to grow in a team that builds its AI skills every day, visit our Careers page.

1. DORA, State of AI-assisted Software Development 2025
2. Stack Overflow, Developer Survey 2026
3. Stack Overflow, Developer Survey 2025
4. Cui et al., The Effects of Generative AI on High-Skilled Work, Management Science
5. METR, 2025 study (arXiv 2507.09089)
6. METR, update of 24 February 2026
7. Anthropic, How AI assistance impacts the formation of coding skills
8. Brynjolfsson, Chandar, Chen, Canaries in the Coal Mine? (Stanford Digital Economy Lab)
9. No Fluff Jobs, IT job market in Poland 2025/2026 (PDF, Polish)
10. Linux Foundation, 2026 State of Tech Talent Europe