In early July, the CEO of one of the world’s largest AI spenders told his own company that agentic development hadn’t accelerated the way his team expected. The admission traveled fast, as these things do, and it landed on a market already primed to hear it. Investors have lost patience with AI stories that lack proof of ROI, and every earnings season adds another round of the same question: is any of this actually working?
It’s a fair question. It’s also aimed at the wrong source.
Narratives move faster than enterprises do. Perception of AI’s trajectory is being set by keynotes, capex guidance, token consumption, and quarterly anecdotes, while the reality is being set inside enterprises by the IT decision makers deploying (or quietly declining to deploy) these tools. When we surveyed those buyers in the inaugural edition of ETR’s monthly AI Tools Pulse, fielded in June, a different picture emerged.
Buyers rate agentic execution far higher than the narrative implies. Technology leaders rated the leading platform, Anthropic Claude, 4.15 of 5 for agentic task execution, with the rest of the field clustered between 3.53 and 3.68. More than half a point separates first from last, and among practitioners using Claude for agentic work, 79% said it performs “well” or “exceptionally.” Not one rated it below “adequate.” That is not the profile of a capability that has stalled.
The value shows up in production, not the pilot. Across platforms, the practitioners who rate agentic capabilities most highly are the ones furthest into deployment. Broad enterprise users consistently score these tools above those still evaluating or piloting. If agentic AI were losing steam, that relationship should run the other way.
Demand isn’t behaving like disappointment. In ETR’s latest Macro Views Survey, 47% of enterprises reported AI spending moderately to substantially exceeding budget, rising to 54% among the Global 2000. Rather than pulling back, most are funding the overage by trimming elsewhere in the IT budget, and only 21% of large enterprises enforce hard caps on consumption. Buyers don’t overspend, uncapped, on things that aren’t working.
The frustrations are real, but specific. Buyers do flag agentic reliability: agents acting outside their instructions, accuracy that’s hard to certify. That is a guardrail problem, and a solvable one. It is a very different diagnosis than “the capability stalled.”
None of this means every vendor wins. The same research shows the gap between the leading tools and the rest widening, and AI budget pressure increasingly spilling into adjacent software categories. The perception gap cuts both ways; some of the loudest AI winners in the narrative look considerably more fragile in buyer data, and vice versa.
Which brings me to how we measure it. For more than fifteen years, ETR’s quarterly Technology Spending Intentions Survey (TSIS) has tracked enterprise spending plans across hundreds of vendors, and it remains the deepest forward-looking read anywhere on how enterprises are allocating to AI. It was quarterly TSIS data that first captured the reordering now playing out among the AI platforms, well before it reached the headlines. But AI also moves between quarters. Model releases can reshuffle the competitive order in weeks, and consumption decisions are made monthly, not annually. Perception gaps open, and close, in the space between quarterly reads.
So, this year we added a complementary layer: the AI Tools Pulse, a standardized monthly survey of IT decision makers across the leading AI platforms, tracking adoption, usage, consumption versus expectations, satisfaction, strategic importance, and real-world use cases. The AI Tools Pulse adds frequency exactly where the market moves fastest. The figures above come from the inaugural June edition. Results from the second edition were released to ETR clients today.
The takeaway travels well beyond any single data point: before accepting an AI narrative, bullish or bearish, ask whether it is coming from the people building the models or the people buying them. Right now, those two groups are telling very different stories. We measure the second one, every month.
If you’d like to see how, reach out to our Service Team.