Two of the most frequently cited Gartner forecasts on agentic AI point in opposite directions, and both are true at the same time. One says AI agents are about to be everywhere in enterprise software. The other says a large share of agentic AI projects now underway will be abandoned within a couple of years. For anyone building a career around this technology, the gap between those two numbers is the most useful data point in the whole conversation.
On the growth side, Gartner has projected that roughly 40% of enterprise applications will include embedded, task-specific AI agents by the end of this year, up from under 5% just two years earlier — a genuinely fast adoption curve by enterprise software standards. On the failure side, the firm has separately forecast that over 40% of agentic AI projects will be scrapped by the end of 2027, citing rising costs, unclear business value, and inadequate risk controls as the main reasons.
The two forecasts aren't actually contradictory once you separate "agents get embedded into software" from "custom agentic AI initiatives succeed." Vendors shipping agent features inside existing products (a scheduling agent inside a CRM, a triage agent inside a support desk) explains most of the adoption curve — that's a relatively low-risk, incremental rollout. The projects most likely to fail are the ambitious, build-it-yourself initiatives: multi-step autonomous agents meant to handle open-ended business processes without much human oversight. Those run into the problems Gartner names directly — governance that wasn't designed for autonomous decision-making, costs that balloon once an agent is calling expensive models repeatedly, and a business case that was never rigorously tested before the build started.
Whichever side of the 40% a given project lands on, a few skills stay valuable either way: writing clear specifications for what an agent is and isn't allowed to do, designing evaluation and monitoring for agent outputs before they reach customers, and knowing how to scope an agentic project narrowly enough that it can actually be shipped and measured. These are the skills that separate the agent projects vendors ship successfully from the ones enterprises cancel after burning a budget on them.
If agentic AI is part of your target role, resist the urge to lead with "I can build an AI agent" — plenty of people can now assemble a demo. Lead instead with evidence that you can scope one responsibly: a project where you defined success metrics up front, built in guardrails, and could show a measurable before-and-after. Hiring managers who've watched a pilot get cancelled are specifically listening for candidates who understand why pilots fail, not just candidates who can describe what agents do.