40% of AI agent projects will be cancelled. Here's why the others will hold.
Gartner predicts four agentic AI projects in ten will be cancelled by the end of 2027. Not because agents don't work: because they're deployed without a plan, without rules and without a human to decide.
In June 2025, Gartner polled more than 3,400 organisations investing in agentic AI. Conclusion: over 40% of projects will be cancelled by the end of 2027. The reasons cited by analyst Anushree Verma fit in one sentence: hype-driven experiments, poorly applied, with no clear value and no governance. Gartner adds a useful detail: of the thousands of vendors calling themselves “agentic”, about 130 genuinely are. The rest are rebranding existing tools.
That number is not an argument against agents. It's the spec for the ones that will last.
A media agent that lasts starts with a written plan. Not a prompt: a document that says what the budget is for, which KPIs matter, which channel mix is targeted, which phasing is acceptable. Without that plan, the agent optimises toward whatever the platform offers — and the platform always offers more spend.
It then has rules set properly, and set once: negative lists, brand exclusions, budget caps, brand-safety rules, ROAS floors. An agent without rules is an intern holding the company card.
Finally, it has a human who decides what to stop, and when.
At mediagentic, these rules apply at three key stages of campaign activity: Plan, Connect, Deliver. More than project framing, it's an operating model we designed for agentic campaigns.
Book a call
Contact us