Selectel: AI doesn't take away jobs, but makes entering the profession significantly harder

AI doesn't cancel work — it changes the rules of access to it. This is the conclusion that Selectel draws by analyzing how artificial intelligence is reshaping the labor market: there aren't fewer vacancies, but entry into the profession is narrowing, requirements are growing, and the price of AI-related skills is skyrocketing. Instead of a scenario where machines massively displace humans, a different picture is taking shape: people are replaced by those who have already learned to work in tandem with models and can faster turn AI into applied results.

Over recent years, an entire layer of new roles has emerged around AI, roles that either didn't exist recently or were niche. At the center of this ecosystem is the LLM/AI Engineer, who doesn't train a model from scratch but assembles a working system from APIs, RAG, tools, and pipelines. Beside him stands the updated MLOps Engineer: his task is now not just to deploy models but to control latency, reliability, and inference costs.

Even higher in demand and salaries is the AI Infrastructure Engineer — a specialist in GPUs, distributed computing, quantization, caching, and other things that directly impact the economics of an AI product. This layer is complemented by the AI Product Manager and AI Interaction Designer: the first balances answer quality, query cost, and user experience, the second designs dialogue logic, system behavior, and how it should fail. A separate tier consists of roles related to data, quality, and safety.

Prompt Engineer from a trendy independent profession gradually transforms into a set of mandatory skills for engineers and product teams. Human-in-the-loop specialists and AI Trainers label data, correct model responses, and essentially create a hidden human layer beneath "automatic intelligence." Synthetic Data Engineer goes further and generates training datasets for rare or underrepresented scenarios.

AI Auditor and AI Risk Specialist check systems for bias, legal risks, and regulatory compliance. In other words, the market isn't just adapting old professions to new tools but assembling a full-fledged production infrastructure around AI. But together with the emergence of new roles, the very principle of hiring changes.

According to data that Selectel cites based on research, up to 66% of companies are reducing the hiring of specialists who will need long retraining, and about 90% of employers are noting the disappearance or radical transformation of entry-level roles. This hits juniors particularly hard: typical code, basic analytics, report preparation, and other routine work are increasingly automated. Companies are less willing to hire someone "for growth" and prefer candidates who already know how to use AI tools in work processes.

Against this backdrop, the market is polarizing: the upper segment with AI engineering, infrastructure, and product management is getting more expensive, the lower end is being squeezed by automation, and mid-level positions find themselves in the middle — some of their tasks disappear while the rest require almost senior-level expertise. In parallel, the premium for AI competencies is growing: having such skills, by some estimates, can add up to 15% to salary. Layoffs haven't gone away, but it's important to interpret them correctly.

Selectel cites an estimate that in 2025 the global tech sector lost about 246 thousand employees, of which about 55 thousand layoffs were in some way connected to AI. In the first months of 2026 the trend continued: about another 40 thousand cuts, and the average intensity increased from 674 to 926 people per day. However, AI is rarely cited as the sole cause of layoffs.

Companies usually speak of restructuring, efficiency gains, and reallocation of resources toward AI-related directions. It's also important that AI's direct contribution to layoffs remains limited: in early 2026 in the US it was linked to only about 7% of cases, and CFOs on average expected workforce reduction of only about 0.4% per year.

This shows that AI often acts as an accelerator of optimization that has already begun rather than as the sole source of crisis. The main conclusion here is not that there is less work, but that the labor market has become significantly more selective. Specialists win who understand model limitations, can calculate inference economics, build reliable pipelines, and take on more complex tasks than before.

What loses is not the profession itself but the old way of entering it — through simple tasks, long ramp-up, and gradual learning on the job. That's why the real competitor today is not AI as such, but the person who has already embedded it into their daily work and learned to get measurable results from it.