End of the AI euphoria? Companies count tokens and hold off on layoffs

Last year’s predictions of mass AI-driven layoffs have not yet matched reality. While 32% of companies worldwide using artificial intelligence models expected to have fewer employees by the end of 2025, only 14% now say they actually reduced their workforces because of AI, according to a McKinsey survey of more than 1,700 companies in 97 countries.

Companies also created fewer new jobs than they had planned. Thirteen percent had expected to increase employment because of AI, but only 8% have decided to create new positions. The most common response has been to keep staffing unchanged, reported by 66% of respondents.

“

After the initial euphoria around artificial intelligence, it turned out that it is not easy to translate the individual productivity gains of employees using AI tools into improved productivity for the company, while the total costs of using artificial intelligence quickly rise with its popularity.

The slower employment impact reflects the more cautious pace of AI deployment and development. About one in five companies said the operating costs of AI models, including token costs, were restricting expansion. In information technology and finance, the share rose to one in four.

“

Although the unit cost of a token has fallen, the number of tokens consumed and generated has grown even faster.

Magdalena Warzybok, a director at Mercer Polska, told Rzeczpospolita that companies had discovered how difficult it was to turn individual productivity gains from AI tools into higher productivity across the business. At the same time, total costs rise quickly as use becomes more widespread.

More companies, including in Poland, are introducing “token counters” that show employees how many tokens they use for particular tasks. Michael Chui, a senior McKinsey expert, said the unit cost of a token had fallen, but the number of tokens consumed and generated had risen faster. Finance and AI executives are increasingly discussing “tokenomics” and the return on AI investment, especially for complex reasoning and autonomous-agent software, where advanced models can be expensive.

“

They have realized that for complex tasks requiring advanced reasoning or software development based on autonomous agents, where the greatest benefits come from frontier models, AI costs are not negligibly low.