AI Deflation in IT: Klarna and IBM Cases Show Why Job Postings Are Up but Pay Growth Is Weak

AI is not killing demand for developers directly, but it is quickly cutting the price of routine engineering work. Job postings are multiplying while pay grows ever more slowly, and for some senior engineers the market has already turned negative. The author calls this AI deflation: businesses get more output from AI tools but do not share the gain with employees through salaries.

According to US data, IT compensation grew just 1.6% in 2025, the worst figure in roughly 15 years. In some segments, senior developers lost about 10% year over year, and mid-level SQL developers lost about 7%. At the other end are AI specialists, whose pay is rising 9–12%, with average compensation for LLM developers reaching 209,000 dollars a year. The market is splitting into two layers: those whose skills AI makes more valuable, and those whose work AI makes cheaper.

The most interesting paradox is that demand is not formally falling. According to Citadel Securities and Indeed, the number of software engineer postings rose about 11% year over year. A couple of years ago that would almost automatically have meant another round of salary competition.

The logic is different now. If development becomes cheaper and faster thanks to code generation, automated tests, documentation and legacy-system analysis, a company does not necessarily hire fewer people. It may hire the same number or even more, but on more modest terms.

The market gets more positions, but the average value of a typical position falls. That is what the deflation label describes: not a collapse in employment, but a falling price for standard work.

This pattern is already visible at large companies. Block carried out major layoffs, and the market took it positively: the company's shares rose. IBM reported billions in operating-cost savings from AI and automation, including sharp cost cuts in HR and IT functions. BCG records an even more telling signal: 93% of executives see AI primarily as a cost-cutting tool, not as a source of new pay premiums.

The business logic is transparent. If AI makes a given activity 15% more efficient, a company either shrinks the team or asks it to handle 15% more tasks. The option of "paying 15% more" barely appears in these models.

The consequences differ by level. Juniors have it hardest: AI eats exactly the starter tasks on which they used to build experience, such as boilerplate code, simple integrations, initial documentation and basic bug triage. Entry to the market becomes narrower and more demanding.

Mid-levels are hit if their value comes down to quickly writing standard CRUD and maintaining predictable services. But strong mid-levels have the best chance to come out ahead: with a good set of AI tools and an understanding of the product, they can cover a volume of work that once required a larger team. Seniors are holding up better for now, but for them too, status no longer protects by itself.

The market increasingly pays not for years in the profession but for the ability to make architectural decisions, reduce risk, lead a team and turn AI into a measurable business result.

The Russian context deserves separate attention. The picture here is so far milder than in the US, but signs of cooling are already visible. According to data cited by the author, the median IT specialist salary in the second half of 2025 was 183,000 rubles and barely changed over the year. In Moscow, growth was about 4%, while Rosstat-measured inflation reached 5.6%, meaning incomes effectively fell in real terms. Across Russia as a whole the market looks uneven: in some places salaries are still rising, in others stagnation is already apparent. So it is too early to speak of proven AI deflation in Russian IT, but a weakening in pay dynamics is already visible.

The main conclusion is simple: AI does not abolish the developer profession, but it changes the structure of value within it. What gets cheaper is not "development in general" but routine, easily scalable work. Roles that require domain expertise, architecture, accountability for results and the ability to build AI into a product or process, rather than just using editor suggestions, get more expensive.

If this logic takes hold, the market could show growing job counts alongside weak pay for a long time. For professionals it means one thing: the winners will not be those who simply write code faster, but those who use AI to do noticeably more for the business than their salary would have bought under the old rules.