Gartner's May 2026 forecast handed the agent industry the number it wanted: spending on purpose-built AI agent software will hit $206.5 billion this year, up roughly 139% from $86.4 billion in 2025, then climb to $376.3 billion in 2027. It's the kind of figure that anchors a keynote slide. It's also, read beside Gartner's other forecasts, a more complicated story than "agents are winning."
| Year | AI agent software spend | Year-over-year |
|---|---|---|
| 2025 | $86.4B | — |
| 2026 | $206.5B | +139% |
| 2027 | $376.3B | +82% |
The number behind the number#
Start with what a spending forecast actually measures. It is gross outlay — dollars that leave a budget — not deployments that work. A market can post 139% growth because a hundred new things succeeded, or because the same fifty things were bought, scrapped, and bought again under a new name. The headline can't tell you which, and with agentic software in 2026 the evidence points hard at the second.
Two of Gartner's own numbers do the pointing. The firm predicts that more than 40% of agentic AI projects will be canceled by the end of 2027 — citing escalating costs, fuzzy business value, and weak risk controls. And in May it cautioned that AI-driven layoffs may create budget room without delivering the returns that justified them. Those are not footnotes to the growth story. They are the growth story, viewed from the cost side.
A 139% spending jump is exactly what churn plus reclassification looks like.
Churn and reclassification#
Churn first. When a project counted in this year's spend gets canceled and a replacement gets funded next year, the replacement is new spend. A high failure rate doesn't suppress the curve — it feeds it. The faster organizations abandon agent deployments and try again, the more the line climbs. A spend total rewards activity, and cancel-and-rebuild is activity.
Reclassification is the quieter inflator. "Purpose-built AI agent software" is a category boundary, and boundaries move toward the budget that's approved. A workflow-automation feature, an RPA bot, a SaaS module with a planning loop bolted on — relabel it "agentic" and it counts. Gartner itself flagged the demand pull when it predicted 40% of enterprise apps would ship task-specific agents by 2026, up from under 5% the year before. Some of that is genuinely new capability. Some of it is the same software wearing the season's word.
Where it sits#
For scale: the agent-software line lives inside Gartner's broader call that total worldwide AI spending reaches $2.59 trillion in 2026, up 47%. Against that, $206.5 billion is a slice — but the fastest-growing slice, and the one where the gap between money spent and value delivered is widest, because it's the layer where agents are still failing in production at rates that make the cancellation forecast plausible.
So take the $206.5 billion for what it is: a measure of how hard enterprises are trying, not of how often it's working. The two are usually correlated. In 2026, for agents specifically, they have come apart — and the honest way to read the chart is with both lines on it. The spending curve climbs because the cancellation curve does. Before you add to either, define the task narrowly, instrument the outcome, and budget for the rebuild — because by Gartner's own math, the average agent project is likelier to be canceled than to quietly succeed.
