The week's most-shared AI stat didn't come from a model benchmark. It came from accounting. KPMG's Global AI Pulse survey for Q2 2026 found that 49% of senior leaders have scaled back or delayed AI agent deployments because operating costs outweighed the benefits — and the number went viral before anyone read past the headline.
But here's the twist most coverage missed: the same survey shows average AI spending holding steady at $188 million, with 79% of leaders still calling AI a top investment priority. The pullback isn't a rejection of AI. It's a billing problem hiding inside an adoption problem.
I went through the full KPMG dataset as reported by Forbes this week to find what's actually driving the cut — and it comes down to something most AI articles never mention: how agents are metered.
The myth vs. reality table
| MYTH | REALITY |
|---|---|
| "AI agents are getting cut because AI failed" | AI spend is flat at $188M and 79% still rank AI a top priority — agents are being redesigned, not abandoned |
| "Agent costs are unpredictable, that's why we cut them" | Costs are predictable — they're just invisible. Only 26% of leaders have real-time visibility into AI spend |
| "A flat subscription means my AI bill is predictable" | Vendors moved to usage-based token billing; agents multiply metered steps on every task |
| "If agents worked, companies wouldn't cut them" | 81% of Asia-Pacific firms report real business value from AI — the cuts are rephasing, not verdicts |
Myth 1: "The AI bubble is bursting — the agent pullback proves it"
The 49% number looks like bubble evidence. Prediction-market traders even ran with it, pushing the odds of an AI bubble burst by year-end to roughly 15%. But the rest of the KPMG data tells a different story.
- AI spend didn't drop. Average AI spending held steady at $188 million per organization.
- Priority didn't drop. 79% of leaders still call AI a top investment priority, up from 74% the previous quarter.
- Everyday use jumped. The share of organizations describing AI as part of everyday work rose to 22% from 13% in Q1 — the largest single-quarter move KPMG has recorded at any stage of its maturity curve.
Companies that pulled back on agents aren't exiting AI. Per KPMG, they're looking at how they will redesign them. That's a budget rephase, not a bubble burst.
Myth 2: "Agent costs are unpredictable — that's why we cut them"
This is the myth that hurts the most, because it's half true. Agent costs DO feel unpredictable. But the survey shows the real problem: almost nobody can see the bill.
In KPMG's companion US pulse survey — 204 leaders at billion-dollar companies — only 26% report full, real-time visibility into what AI costs to run at scale. In the global survey, a third of leaders cite limited understanding of AI cost structures, including how token pricing works, as a barrier to deploying agents.
So the sequence goes: companies deploy agents, the bill arrives opaque, finance panics, agents get cut. The fix isn't fewer agents. It's cost meters, token education for leadership, and cost reviews built into the AI approval process — exactly what KPMG's analysts recommend.
Myth 3: "A flat subscription means my AI bill is predictable"
Here's the part that's genuinely new this year: vendors have shifted from flat subscriptions to usage-based pricing billed in tokens. A token is a small chunk of text, roughly a word fragment. Every question an AI system reads, every answer it writes, and every step it takes consumes tokens — companies pay per unit, the way they pay for electricity.
Agents changed the math because they work differently than chatbots. A chatbot answers one question and stops. An agent runs long tasks, calls other software, and checks its own work — and every one of those steps is metered. One tool-heavy agent session can consume what a month of chat never would.
The example that made this real: when GitHub Copilot moved to usage-based billing on June 1, a Visual Studio Magazine writer tracked his first day under the new meter and projected a $180 monthly bill on a plan that had been a flat $10 — driven by a single long, tool-heavy session.
That's the billing math behind the 49%.
Myth 4: "If agents were valuable, companies would never cut them"
The loudest takeaway from the 49% stat is that agents don't work. The data says the opposite.
In Asia-Pacific, 81% of companies report AI already delivering meaningful business value, up from 69% three months earlier. And the organizations cutting agents are mostly clearing room to scale what works tomorrow.
Sandy Carter, the Forbes contributor who analyzed the data, put it bluntly: "49% of enterprises scaled back when the cost was greater than the benefit but only 26% have a real time view of what AI costs to run."
You can't manage what you can't measure — and 74% of billion-dollar companies can't see their AI bill in real time.
What this means for you
If you use AI agents at work — coding assistants, research agents, automated workflows — this pullback is going to touch you whether your company is one of the 49% or not:
- Expect usage dashboards. If your employer is cutting agents, they'll want per-user, per-task token visibility before approving anything new.
- Learn token economics. Chat queries cost a few thousand tokens; agent loops can burn hundreds of thousands per session. One agent session can cost 20-50x a chat query.
- Ask about caps before you build. The companies that kept agents running are the ones that capped spend per team and gated approvals before scaling — the same playbook Rippling used to fix its AI token bill.
- Price agents like infrastructure, not software. Flat-seat pricing is disappearing for agent-heavy tools. Budget per task, not per user.
Key Statistics
- 49% of senior leaders scaled back AI agent deployments because operating costs outweighed benefits — KPMG Global AI Pulse Q2 2026, 2,145 leaders across 20 countries
- 79% still rank AI a top investment priority (up from 74%), with average AI spending steady at $188 million — same survey
- 26% of leaders at billion-dollar US companies have real-time visibility into AI costs; a third cite weak understanding of token pricing as a deployment barrier — KPMG US pulse survey
- 22% of organizations now describe AI as part of everyday work, up from 13% in Q1 — the largest single-quarter jump on KPMG's maturity curve
- 81% of Asia-Pacific companies report AI delivering meaningful business value, up from 69% three months earlier
- GitHub Copilot's June 1 switch to usage-based billing produced a projected $180/month bill for one developer on a formerly $10 flat plan
Frequently Asked Questions
Q: Does the 49% pullback mean AI agents are a failed technology?
A: No. The same survey shows AI spending flat at $188 million and 79% of leaders still prioritizing AI. The cuts are about ROI discipline and metered billing, not capability. Most companies that pulled back are redesigning their agents, not abandoning them.
Q: Why are agents so much more expensive than chatbots?
A: Chatbots answer one question and stop. Agents run long tasks, call other software, and check their own work — and every step is metered by the token. One tool-heavy agent session can cost more than a month of chat usage.
Q: I use Copilot or Claude Code. Will my bill explode?
A: Possibly, if you run long agent sessions. GitHub Copilot moved to usage-based billing on June 1 — one developer projected a $180 monthly bill on a plan that was $10 flat. Check your usage dashboard and set spending alerts before running multi-hour agent tasks.
Q: How do companies actually fix the AI cost problem?
A: KPMG's analysts recommend cost meters, token-pricing education for leadership, and cost reviews inside the AI approval process. Only 26% of leaders currently have real-time visibility into AI costs — fixing that visibility is step one.
Q: Is the AI bubble bursting?
A: Prediction markets put the odds of a bubble burst by year-end at roughly 15% — hardly a foregone conclusion. KPMG's data shows budgets following results instead of promises, which is what a maturing market looks like, not a crash.
The bottom line
The 49% number isn't a verdict on AI agents. It's a receipt.
As Carter put it: "The bill came due. Reading it carefully is not a crash. It is AI agents reaching adulthood." The companies that win the next phase won't be the ones with the best models — they'll be the ones who can see their token bill in real time and budget agents like infrastructure instead of software.
Some links in this article may be affiliate links.