From Enthusiasm to Economic Calculation
Over the past two years, the market has been driven by a logic that analysts have sarcastically dubbed 鈥渢okenmaxxing鈥�: the more data, computing power and prompts, the better. Budgets flowed freely because no one wanted to be left behind. Many executives assumed that higher levels of token usage would translate into greater automation of business processes. Automation, in turn, promised significant savings. By the summer of 2026, however, the mood had changed noticeably. Business leaders stopped investing money without a realistic prospect of returns, while providers of large language models, once symbols of limitless growth, began preparing for stock market listings.
This is a classic stage in the maturation of any technology. First comes fascination, then an investment frenzy and finally a period of sober assessment. The internet went through a similar process at the beginning of the century when the dot-com bubble burst. AI is now entering its own 鈥渟how me the numbers鈥� phase.
The 95 Per Cent That Vanishes Without a Trace
The mentioned above MIT report has been one of the most sobering assessments of recent months. Researchers analysed 300 implementations, conducted 150 interviews with business leaders and surveyed 350 employees. Their conclusion? Only a handful of projects had a genuine impact on a company鈥檚 financial performance. The rest became trapped somewhere between colourful presentations and the everyday use of tools.
Interestingly, the problem was not the technology itself. Companies most often directed funding towards areas where AI can clearly create value, namely marketing and sales departments. At the same time, the research shows that the highest returns came from less glamorous back-office automation. It also revealed that solutions purchased from specialised providers delivered results in roughly two out of three cases, whereas internally developed solutions succeeded three times less frequently. In other words, the issue was not the technology but the way it was implemented.
The Revolution of Affordability
A second drama is unfolding in the background: a sharp decline in costs. New models, many of them originating in Asia, offer performance comparable to that of market leaders at a fraction of the price. Unveiled at the turn of May and June 2026, the MiniMax M3 model uses a new architecture capable of reducing computing costs by as much as twentyfold, with pricing starting at just a few dozen cents per million tokens.
For businesses, this changes the rules of the game. Not long ago, advanced AI resembled an airport taxi: convenient, but financially unsustainable for everyday use. Today, it is becoming more like a public transport ticket: inexpensive, widely available and accessible to everyone
Why 40 Per Cent of Projects Will End Up in the Bin
AI agents, systems capable not only of responding but also of carrying out tasks independently, have been one of the industry鈥檚 most discussed topics since 2025. Here, however, comes a warning that should keep management boards awake at night. According to analyst firm Gartner, more than 40 per cent of AI agent-based projects will be cancelled by the end of 2027 due to rising costs and unclear business value.
This is effectively a sentence on implementations driven by fashion rather than need. Many companies build AI agents in the same way that people buy fashionable gym equipment, with great enthusiasm that evaporates after three weeks. The greatest risk today is not that AI will fail technically, but that it will be implemented without a clear purpose or business rationale.
Not a Tool, but a Way of Working
For organisations, the lesson is clear: competitive advantage will belong not to those with the greatest number of tools, but to those capable of changing the way they work. Artificial intelligence does not replace strategy; it ruthlessly exposes its quality. Where processes are chaotic, AI will amplify that chaos. Where they are well designed, AI can strengthen and enhance them.
That is why businesses should now ask themselves less glamorous, but far more important questions than 鈥淲hich model should we choose?鈥�. Which specific process do we want to improve, and how will we measure success? Are our employees genuinely capable of using these tools in their daily work, or have they only seen them demonstrated in presentations? AI training alone is not enough if people return to old habits as soon as the course ends. AI competencies are becoming a fundamental element of professional effectiveness, as essential today as basic computer literacy.
Maturity Instead of Hype
The end of the 鈥渢oken-burning era鈥� is good news. It means that artificial intelligence is no longer merely a topic for conference slides but is becoming a tool that simply has to generate value. If that is the case, success will belong not to those who spend the most, but to those who best combine technology with the skills of the people who know how to use it.
Author: Dr Dominik Skowro艅ski
Edit: