Leadership1 publisher3 min readPublished
Agent budgets head for $65.5bn next year with many companies unable to measure what each agent returns
Gartner expects spending on AI agents and assistants to more than double to $65.5bn next year, while many firms cannot measure the return. Splunk executives who described that gap were selling a tool to close it, and the tool leaves buyers to define what an agent's work is worth.
The Board Room · Leadership desk
Drafted by a language model from the sources cited here and checked against its claim ledger before publication. How we use AISend a correction
What happened
- Gartner puts total worldwide AI spending at $2.7 trillion this year, up 49.5% on last year, rising to $3.64 trillion in 2027.
- An employee using a coding agent can set off a chain of model and tool calls without knowing what each costs, a pattern Splunk's Vikram Chatterji said ordinary IT cost controls were not built for.
- Splunk Security chief John Morgan cited statistics showing the top 1% of companies spending upwards of $7,000 a month per employee and the next 10% about $700.
- Chatterji co-founded Galileo, the AI evaluation firm Cisco finished buying on 22 May, whose technology now underpins Splunk's new agent-monitoring product.
Compiled by The Board RoomSomething wrong?How this is made
Why it matters
- decision With the agent line growing about 124% next year, companies setting that budget choose between a flat cap that needs no data and value-and-risk limits that need per-agent baselines first.
- exposure Employees can commit spend through chains of model and tool calls they never see priced, so cost builds up outside the approval steps IT finance already runs.
- cost Morgan expects agents to raise security spending, and judging it against the organisation's risk profile means the security line needs its own return measure, separate from productivity.
Inside Gartner's forecast, the agent line is growing much faster than the total. Going from $29.2bn to $65.5bn [2] is growth of about 124% [1]. All AI spending grows roughly 35% between this year and 2027 [1][2]. Agents and assistants are still a small slice: about 1.1% of AI spending this year and 1.8% next [3].
Where the money goes affects how it can be counted. "There is also a pattern when it comes to cost. Most of the cost right now is through third-party coding agents," said Vikram Chatterji, Splunk's senior director of product management, at the company's .conf26 event in Denver [4][6]. The buyer did not write those tools [4]. Splunk launched its response at the same event, with a tokenomics feature covering AI spending and coding-agent usage [12]. It pulls telemetry from first-party agents and third-party tools into one view [13].
Over a year, John Morgan's figures come to $84,000 a head for the top 1% of companies and $8,400 for the next 10%, a tenfold gap [4]. The article does not say where his statistics came from [7]. With a gap that wide, one limit cannot suit every company. "You can't have a single throttle of saying if it's above $5, stop it," Chatterji said [8].
The trade-off is between a simple control and one that needs a definition of value. Chatterji's alternative weighs the value generated and the risk of each agent interaction [9]. A cap does not ask what a good outcome is worth. The weighted control needs that answer for every agent. Hao Yang, Splunk's head of AI, said it has to be given in terms business leaders understand, beyond tokens consumed or lines of code generated [6][10]. Chatterji described what comes once the numbers exist: "If the cost is high and the ROI is low, the third piece becomes 'what is the action you need to take', which gets into this notion of how do you optimise your entire agentic system overall" [16].
A skeptic would say the panel describing the measurement gap was selling the fix. That holds for the framing. The claim that many enterprises cannot measure return is iTWire's account of a vendor panel [3]. The growth figures come from Gartner, not Splunk [2].
Security brings a return that is harder to count. "Agents are going to help with figuring out threat environments that are specific to your environment, but it will also increase your overall security cost," Morgan said [14].
This quarter's decision determines which control a company can use next year. A company that defines value per agent before the budget is set will have a baseline when the bills arrive. One that starts afterwards will be measuring spend it has already committed. I think the cost half is easier to build, because usage telemetry can be collected from the tools [13]. The value half is internal work, and Splunk's own design leaves each customer to define it [17].
What to watch
- A named, sourced dataset on per-employee agent spend that tests the $7,000 and $700 monthly figures Morgan cited.
- Whether Gartner's next update moves the $65.5bn agents-and-assistants forecast or splits coding agents out of it.
- Whether companies using tools like Splunk's publish the per-agent value measures those tools leave them to define.