An agent can keep spending after the person who started it has stepped away. Menlo Ventures’ new enterprise AI report makes that operating pattern a budget question.

In its November 3 report, Menlo estimates enterprise AI deployment at a $177 billion annualized run rate by September 2026, compared with a reclassified $33 billion for 2025. These are modeled market figures. The methodology says revenue is annualized, so the headline should not be read as cash already spent during the calendar year.

The report draws on nearly 400 U.S. enterprise decision-makers responsible for more than $4 billion in AI spending, surveyed by Morning Consult in September. Market sizing also uses public filings and third-party reports. Menlo is an investor in several companies discussed, including Anthropic and OpenRouter.

One finding deserves a closer look before it enters a planning slide. Survey respondents reported average spending at 4.7 times the prior year’s level, against a planned 1.9 times. The median was lower: 2.2 times actual against 1.8 times planned. A few very large users can pull the average away from the experience of a typical buyer.

For an enterprise, that creates a reason to change the unit of budgeting. A seat count limits how many people can open a tool. An agent workflow can make repeated model calls, invoke paid services and retry failed steps within one person’s request. The useful comparison is the cost of an accepted result, with those attempts included.

Give the workflow owner a usage limit and a record of completed work. Include the staff time spent checking exceptions and repairing results. A growing market can explain why bills are rising; the budget still needs evidence of what those bills bought.