The global market for AI agents is growing by close to 50% a year and is expected to rise from $10.9 billion in 2026 to $182.9 billion by 2033, according to Grand View Research. In Luxembourg, 59% of the organisations surveyed by FEDIL, the Luxembourg AI Factory and Luxinnovation have already deployed generative AI company-wide or on concrete use cases in production. The movement is real, it is fast, and it will not slow down to wait for latecomers. Yet Gartner also predicts that more than 40% of agentic AI projects will be cancelled by the end of 2027. This opinion piece explains why the two figures do not contradict each other, and what they mean for a company that wants to deploy AI agents without swelling the wrong statistic.
Why is the window closing faster than it opened?
The momentum is no longer a hypothesis. A large majority of respondents (88%) in the survey run by FEDIL, the Luxembourg AI Factory and Luxinnovation among 136 professionals rank productivity gains first among their reasons for adopting AI, and Luxembourg companies have already moved from awareness to integration. In France, Bpifrance Le Lab found as early as June 2025, among 1,209 heads of SMEs and mid-sized companies, that 58% of them see AI as a matter of survival within three to five years, a proportion that doubled in a year. Intent is no longer the problem. The problem is knowing who will turn it into results.
Two deadlines now structure the calendar. The EU AI Act entered its enforcement phase on 2 August 2026: authorities can now fine prohibited practices up to €35 million or 7% of worldwide turnover, and breaches of transparency obligations up to €15 million or 3%. The obligations specific to high-risk systems, including recruitment, education and access to essential services, will apply from 2 December 2027, following the postponement decided by the European simplification package adopted in June 2026: that is precisely the horizon at which Gartner places the sorting between the agent projects that will hold and the rest. On the skills side, finally, ManpowerGroup reports in its 2026 survey of 39,000 employers across 41 countries that AI skills have moved, for the first time, to the top of global talent shortages, ahead of traditional engineering and IT. Recruiting an in-house team able to build, govern and maintain AI agents now takes longer than waiting for the next model.
Companies that are getting ahead no longer ask whether they will use AI agents. They ask how to do it without ending up in the wrong statistic.
Why do most agentic AI projects fail despite this momentum?
Wanting to automate is no guarantee of successful automation. Gartner predicts that more than 40% of the agentic AI projects launched today will be cancelled by the end of 2027, before they even reach production at scale. The causes it identifies have little to do with the technology itself: costs that spiral because nobody priced the integration with existing systems, business value never defined before the project started, risk controls missing once the agent is able to act on its own on sensitive data.
Add to that what Gartner calls agent washing: most of the providers selling AI agents today are in fact selling chatbots, assistants or conventional automations renamed for the occasion. Of the thousands of vendors claiming the capability, the firm estimates that only around 130 genuinely have it. A company that goes it alone, or hands the subject to the first provider that comes along, statistically runs a greater risk of joining the 40% of failures than of joining those that gain a real advantage.
The question is therefore no longer whether to go ahead. The question is who to go ahead with. And that decision is climbing ever higher in the organisation chart: 72% of executives now describe themselves as their company's main AI decision-maker.
AIxH's view
A poorly scoped AI agent is an operational risk: access to sensitive data, actions taken without supervision, dependence on a provider who understands neither your business nor its own tool. That is precisely what our AI audit and integration in Luxembourg service is built to prevent, from initial scoping to the deployment of agents and RAG systems. We do not sell a generic AI agent renamed for the occasion. We define the business value before the first prompt, we integrate the agent with your existing systems rather than alongside them, and we keep control of risk governance from day one through to production. The window mentioned at the start remains open, but it does not reward speed alone. It rewards those who get the right people around them before they rush. An AI maturity audit usually settles the question in a few weeks rather than several seasons of deliberation.