Agentic AI is the phrase every vendor deck and every board meeting is chasing this year. It is also, for most of the companies rushing to deploy it, a mistake they have not noticed yet. I say this as someone who thinks agents are genuinely useful. That is exactly why the current stampede worries me.
The numbers are already whispering
Gartner expects more than 40 percent of agentic AI projects to be cancelled by 2027. One widely cited study found only about a quarter of AI initiatives delivered the ROI companies expected. Those are not the numbers of a technology that is failing. They are the numbers of a technology being deployed badly, by people who bought the demo and skipped the design.
Why the failures rhyme
The doomed deployments share a shape. An agent gets pointed at a vague, sprawling process nobody fully understands. It has no clear definition of success, so nobody can tell if it is working. It is wired into systems it does not really comprehend. And it is given autonomy over actions that are hard to undo. Then everyone acts surprised when it makes confident, expensive mistakes.
Handing an agent a messy process does not clean up the process. It just automates the mess and speeds up the regret.
The uncomfortable diagnosis
Here is the part nobody in the meeting wants to say: if your process is too vague for a competent new hire to run from a written description, it is too vague for an agent. The agent did not fail. It exposed that you never actually understood the workflow you asked it to automate. Agents are a mirror, and a lot of organizations do not like the reflection.
What the good deployments do differently
- They pick one narrow, well-understood task, not "the company."
- They keep the actions reversible and a human on the trigger for anything that is not.
- They define success in advance and measure against a manual baseline.
- They put hard limits on the loop: max steps, max spend, max retries.
So what do I actually think
I think agents will matter enormously, and I think most of the ones shipping today will be quietly killed within eighteen months. Both of those are true. The technology is ready for narrow, well-scoped, supervised work and nowhere near ready for the autonomous everything the marketing implies. The companies that win will be the ones patient enough to start small. The ones chasing the buzzword will fund a very educational round of cancelled projects, and the rest of us will learn from them for free.