Prosus built 60,000 AI agents and is scaling it back to 5,000.
KPMG surveyed 1,200 companies: only 8 percent sees a clear return. The costs of AI agents are high and unpredictable — but so is the return. Whoever makes the distinction has control over both.
By Anthony Raaijmakers
Prosus built 60,000 AI agents and is scaling it back to 5,000.
Het Financieele Dagblad wrote about it last week. KPMG surveyed 1,200 companies: only 8 percent sees a clear return. 80 percent of multinationals deploying AI agents conducted a round of layoffs. Gartner analyst Helen Poitevin is clear: "Layoffs create budget space, but they don't generate profit." Token consumption is now three times higher than a year ago.
What we specifically see with AI agents: costs don't scale like regular software. At Amsterdam-based Mews, panic briefly set in when the real Anthropic bills arrived. A new employee burned through €1,000 in tokens in a single day. But built something with it that a team would normally take months to do.
The costs are high and unpredictable, but so is the return. At OneMinded, we therefore always ask that question first: what is the problem, and what is the most future-proof solution? Sometimes that's an AI agent. Sometimes it's a process built with AI, but running without it. Whoever makes that distinction has control over both.