Czech developers are changing jobs less but building more

Look at the market as a whole and hiring looks flat, roughly covering what leaves. That average hides the real story, so we split the pool by company. We track 1,327 Czech tech employers. Most are small, so we looked at the 320 with a team big enough to read, 30 or more tracked developers. That is a quarter of the companies but three quarters of every developer we see.

Those 320 break into four very different piles. Only 10% grew their engineering team over the past year. Another 14% roughly replaced what they lost. Nearly half, 46%, barely moved. And 29% ended the year smaller, refilling less than 70% of who walked out.

So a flat market is not a market of steady teams. It is a big frozen middle, a thin band still growing, and almost a third quietly getting smaller. For sourcing, that last column is the one to read. Shrinking teams are where strong developers become available.

Developers are leaving their jobs later than they used to, and the shift is recent. For a decade the typical stint before a move barely changed. From 2015 through 2024 it sat between 22 and 27 months, a little under two and a half years, through booms and busts alike. Then it broke. In 2025 the median exit came at 31 months, and in the first half of 2026 at 33. A pattern that held for ten years moved sharply in eighteen months.

The people still in their jobs tell the same story. Their median time in place has risen from three years in 2021 to almost five now, and because the longest stayers never appear as an exit at all, that number understates the shift.

Both point one way. When a market cools, the pull of a better offer weakens and the risk of a move rises, so people stay, and stay longer. The data cannot separate the developer who stays by preference from the one who sees fewer reasons to leave. What it shows is a workforce that is settling in rather than circulating.

What Czech developers build in the open has shifted, faster than the picture of a stable stack suggests. The clearest move is TypeScript. In 2018 it was a rounding error, on 3% of active public repositories. By 2025 it was on 14%, more than fourfold, as the web moved off plain JavaScript onto typed JavaScript. Python climbed too, from 12% to 18%, holding its place as the most used single language across backend, data and now AI.

The other side of the ledger is the old guard. Java more than halved in public code, from 12% to 5%, and PHP did the same, from about 7% to 3%. One honest note on that. Much of the Java written in this country lives in private company repositories that this cannot see, so its public retreat overstates the real decline. Both still have plenty of working developers.

But in the open, the centre of gravity has moved to Python and the TypeScript-led web, with a receding old guard behind. And a language is not a job. This counts repositories by the year they were last active, so it maps where public work concentrates now, not every line written in private.

Most developers do little in public, and that has not changed. But the open source scene, the developers who actually contribute in the open, is unmistakably busier. We track this closely, day by day, across 22 million daily activity records.

Take the roughly 8,850 developers who made at least 100 public contributions over the past year and follow that same group throughout. Their median output rose from 70 contributions a quarter to 119, up 70%, and it rose across the whole distribution, at the 25th, 50th and 75th percentiles alike. This is not a handful of prolific accounts pulling up an average. The entire active scene moved up together.

Two honest bounds on the claim. This is public work, so it speaks to open source, not the private commercial code most people write. And our public record starts in 2025, so it is a clear one-year trend, not a decade. The timing sits right on the arrival of AI coding tools, though this data shows the rise, not the reason. A busier public codebase, at the same time the job market went quiet.

We can see this across more than 200,000 public repositories built by Czech developers. For all the noise about AI, explicit AI and machine learning work is still a small share of what they build in the open. Among the repositories active over the past year, about one in twenty five is clearly AI or ML.

Two things stand out. It is growing. That share sat near 2% from 2021 through 2022, then climbed to 3.6% in 2024 and 4.7% in 2025, more than doubling over the period generative AI went mainstream. And it draws attention out of proportion to its size. AI and ML repositories are under 4% of the active public pool yet collect more than 9% of all the stars, close to three times the attention per project of everything else.

Two things bound this, and both are scope. It is public work, so it misses the AI that companies build in private, which is likely a lot. And AI repos are not always labelled as such, so keyword detection undercounts them. Both push the same way: the real share is higher than what we can see.

Even so, the honest read is not that everyone pivoted to AI. Most public code is still ordinary software. But the corner that is AI is expanding fast, and it is where the crowd is looking.

Fewer new developers are entering the Czech market. In 2021 about a quarter of active tracked developers were within three years of their first job. By 2025 it was closer to one in six, and the early read on 2026 is lower still. This is not a sudden shift. The share has slid every year since before the pandemic, the mark of a market that has been taking in proportionally fewer newcomers for years, not one that flipped a switch when AI arrived.

A note on what junior means here, because the word carries a lot. We use one thing only: time in the field, someone within three years of their first recorded job. We would have gone by title, but in Czech tech almost nobody calls themselves junior. Only about 2% of developers carry a junior, intern or graduate title, and that barely moves, though the field plainly has more newcomers than that. People are “Software Engineer” from their first week. So titles tell us nothing here, and years since entry is the one piece of seniority we can measure cleanly across the whole market. Real seniority is far more than a count of years, and we are not pretending otherwise. This is the part we can see.

We are careful with the exact size. New entrants are the slowest to surface in the data, and coverage was thinner years ago when fewer developers were visible online. Both can steepen the early years, so read the direction, not a precise percentage. The direction is not in doubt. As teams lean out, entry level is the first hiring they pause and the last they restart. For anyone starting out, the door is narrower than it was.