GARTNERPREDICTS40%OFAGENTICAIPROJECTSWILLBECANCELLEDBY2027
The CES festival just happened last week in Las Vegas. Although we weren’t there, it’s not difficult to admit that even from across the world, the future feels as if it’s here already. Jensen Huang standing beneath a wall of dancing robots declaring that "the ChatGPT moment for physical AI is here."
Humanoids shadowboxing, autonomous vehicles promising to revolutionise how we move through cities. A robot folded laundry with surgical precision during Huang’s speech. Ten trillion dollars of computing, he said, is being modernised. The agentic system is becoming the interface to everything.
Hearing all of this hype, this excitement, it’s important to remember a striking figure from a prediction made by Gartner last summer: 40 percent. That’s how many of these agentic AI projects are expected to be cancelled in the next two years.
The reasons they cite are painfully familiar to anyone who has watched technology hype cycles before: escalating costs, unclear business value, inadequate risk controls. But if those are just symptoms, what is the real reason?
Here is something that shocked us back when we first read it. Of the thousands of companies now marketing themselves as "agentic AI" vendors, Gartner estimates that only about 130 are actually selling what they claim to sell.
The rest are engaged in what analysts have started calling "agent washing." It is a similar pattern to what we’ve seen before. Remember when every SaaS product suddenly became "cloud native" even if it was just hosted on someone else's server? Remember when basic if-then automation became "AI powered" overnight?
It seems that when a promising technology emerges, expectations inflate. And vendors scramble to relabel whatever they already have, hoping nobody looks too closely at what it actually is.
Call recording features become "transcription agents." CRM integrations become "activity mapping agents."
Organisations are spending money, launching initiatives, celebrating pilot programs, and watching them quietly die before ever reaching production. At CES 2026, what was "absolutely unimaginable" just a few years ago seemed trivial. Both sides of the coin are true, and that is precisely the problem.
The technology has advanced remarkably. What large language models can do today genuinely was unimaginable five years ago. But the gap between what AI can do in a controlled demonstration and what it can do inside the messy reality of an actual business has never been wider. Recordings of the laundry folding robot have become quite viral. On stage, it worked perfectly (albeit slowly). But the demo did not show what happens when the towel is wrinkled, or when the lighting changes. When someone left a sock tangled in the sheets. In summary, when the robot encounters something it has never seen before.
Demos assume clean data, stable environments, and forgiving failure modes. Enterprises have none of those things.
Last October, Deloitte published the results of a survey they conducted with around 1,800 executives across Europe and the Middle East. They asked a simple question: how long does it take to see a return on AI investments?
Only 6% said they achieved payback in under a year. The majority answered 2 to 4 years - a period dramatically longer than the seven to twelve months that executives normally expect from technology investments.
Meanwhile, a separate study from Kyndryl found that 61% of CEOs feel more pressure to prove AI ROI now than they did a year ago. You can feel the tension building. Boards want results. Vendors promise transformation. Budgets have been allocated. Announcements have been made. And somewhere in the middle, implementation teams are discovering that "agentic" does not mean "automatic." One analyst at CES summarized it perfectly: “CES loves autonomy, but enterprises love accountability. And those two values collide fast”.
After CES 2026, a different light shines on the Gartner prediction.
The projects that will be cancelled are not failures of ambition. They are failures of alignment. Organizations chasing agentic AI because the competition announced an initiative, not because they identified a specific problem worth solving. Teams deploying tools that cannot access their own data, cannot integrate with their own systems, and cannot adapt to their own workflows. Executives approving budgets based on demos that bore no resemblance to operational reality. The successful 60% will not be the companies with the biggest budgets. They’ll be the ones who understand this:
AI agents operate within systems. And the quality of those systems determines everything.
Don’t get the foundations wrong, or even the most sophisticated agent becomes an expensive way to create chaos.
Perhaps the most interesting thing about CES 2026 was what changed in the air. We’ve read multiple opinions noting that the atmosphere felt different this year. Less breathless hype.
AI has become default infrastructure. It is baked into nearly everything on the show floor. And when something is everywhere, it stops being a differentiator. The question shifts from "do you have AI?" to "does your AI actually work?"
Rather than pessimistic, Gartner's prediction is clarifying.
The technology is real and the potential genuine. But the path from that stage in Las Vegas to actual value creation will require the hard, unglamorous work of making sophisticated tools function within complex organisations. The ChatGPT moment for physical AI may indeed be here. But for most of us, the harder work is just beginning.