Published on the 25/06/2025 | Written by Heather Wright

While agentic projects get cancelled…

AI is not doing its job and is further overloading humans and needs to stop bothering us according to Erick Brethenoux.

It’s a perhaps unusual statement from someone who makes his living as lead AI researcher at Gartner, but Brethenoux says AI is failing us today and we need to be clearer about what AI’s role is: To free us and simplify our lives.

“In many cases we studied, AI distracts. It steals away from the value,”

“The future of AI is up to us, not to AI or AGI (artificial general intelligence) or all the other things being talked about,” he says.

He has a particular issue with the influx of to-do lists and suggestions AI is creating. Most of us now receive summaries from meetings, along with a list of follow-up tasks.

“Do you suddenly have more time to read those summaries? Do you suddenly have more time to do more to-dos? I don’t. And yet five out of the 10 tasks that the technology is nice enough to tell me to do, can be automated.

“Why in the world are you bothering me with something that can be automated in the first place?”

Gartner analysis has shown the biggest value with AI comes when the technology is used by low experienced workers doing low complexity tasks, or by highly experienced workers doing high complexity tasks.

A new worker coming into the workforce can benefit from AI providing the knowledge from existing workers, enabling them to reach a higher performance levels, faster – so long as the task complexity is low. On the flip side, those who are very experienced in their work can quickly vet out whether what they are getting from AI is worth it or not and use AI to augment what they are doing in high complexity tasks.

“In that case you are in the zone of productivity and AI provides value. [But] in many cases we studied, AI distracts. It steals away from the value,” he says.

Get it right, however, and he acknowledges that there can be substantial returns.

Mitsui Chemicals used AI technology and GenAI techniques to discover 160 new materials in one year, adding US$1.2 billion in new revenue.

“So don’t get me wrong, AI, in some cases, has that capability [to provide substantial returns]. But not always.”

He believes AI’s role is ‘human-first’ – taking away things that we do on a regular basis but are tired of doing, and helping augment and accelerate what we are doing – without adding to the overload of alerts workers are already experiencing.

He cited the example of a windfarm in the North Sea, which deploys a fleet of 20 drones to analyse the different offshore wind turbines taking multi-spectral analysis images to identify issues.

The drones return to base and ‘talk to each other’ to verify and validate observations and whether something needs to be pushed to a human operator.

“Ninety percent of the time, there is nothing to report. Leave the poor operators alone.”

But when something needs to be done, the system gathers the facts, prepares reports which are sent to an operator to verify.

“Those solutions already exist today. Each of those drones are an agent. They have a different tasks to do, the ability to communicate with their peers and have a very simple goal in mind – to do what they are doing.

“The idea then is to use AI to protect the users… and only penetrate the sphere of the user if they have to. Why are we sending that report every Monday to the sales managers if 90 percent of the time there is nothing to look at?”

Send the report when there is something different. Leave the human alone the rest of the time, enabling them to do other tasks.

In another example, he noted Vizient which before deploying AI asked 1,000 employees what tasks they hated doing on a daily basis. Using that they automated the most complained about jobs.

“How do you think it went? Instant acceptance, instant adoption, zero change management.”

Vizient is now into a second wave of use cases, with employees suggesting other uses for the technology.

Brethenoux says AI will have done its job when it is a, mostly silent, trusted partner giving us more time to think, thoughtfully decide what to do with our day and act and interact.

“But it is up to us to be able to do this. Let’s use AI to be more human.”

Brethenoux also had a bone to pick with AI agents – or more specifically the idea of agents being a new nirvana inextricably tied to genAI.

“Agents are something vendors are pushing hard. But they are not new. We’ve had AI agents in the world for at least 30 years. Most of the agents we have today are in the IoT domain with instrument AI at the edge in sensors, devices and drones, things like that. Most what we do then uses AI agents.”

And on the aspect of genAI and agents: “Agents and generative AI have nothing in common. They are two completely separate fields.”

While agents can use generative AI models, in most instances they don’t, instead using rule-based systems in combination with an optimisation engine, graph networks connected to machine learning models, or a combination of techniques.

“Ninety-nine percent of the AI agents roaming the world today don’t use generative AI. Because they don’t need too.”

His comments come as Gartner research forecasts that more than 40 percent of agentic AI projects will be cancelled by the end of 2027, due ot escalating costs, unclear business value or inadequate risk controls. The research notes that many initiatives are driven by hype and says only about 130 of the thousands of agentic AI vendors are real.

Brethenoux warned companies to beware of that market pressure.

“A lot of the things you develop today don’t need agent capabilities. It is expensive, it is going to be complicated and different from what you do now.”

And while in some instances they may be very helpful, in the majority of instances, that isn’t the case. Companies will need to tread carefully in identifying solutions where AI agents can really make a difference given their intrinsic capabilities.

And where agents do make sense, prepare for software engineering challenges, he warns. But that’s a whole different story.