KPMG survey: 62 percent of large US companies are building or deploying AI agents

KPMG's third-quarter 2026 AI survey finds the share of companies working on multi-agent systems rose from 6 percent to 25 percent in one quarter.

FUFerhat UğurluYazSis TeamPublished: September 24, 20262 min read
In short

In its quarterly AI survey published on September 24, 2026, KPMG said 62 percent of US companies with annual revenue of 1 billion dollars or more are building, developing or rolling out AI agents, up from 53 percent the quarter before. The share working on multi-agent systems rose from 6 percent to 25 percent.

Audit and advisory firm KPMG published the results of its third-quarter 2026 AI Pulse survey in New York on September 24, 2026. Between July 24 and August 25, 2026, KPMG polled 314 senior executives and business leaders at US companies whose yearly revenue is at least 1 billion dollars.

  • 62 percent say they are building, developing or rolling out AI agents, against 53 percent in the previous quarter. The share developing or implementing multi-agent systems climbed to 25 percent, from 6 percent in each of the two prior quarters.
  • 44 percent now say AI is widely used across their workforce, compared with 23 percent a quarter earlier and 10 percent a year before. Nearly 6 in 10 leaders say they see measurable business value.
  • 74 percent include cost reviews in AI approvals (61 percent last quarter), 43 percent use usage or token budgets, and 49 percent have defined high-risk cases where agents cannot make decisions on their own.
  • In a separate KPMG poll of 100 banking leaders, 80 percent named data readiness and access as the top obstacle to deploying agents.

Why it matters

The survey covers large US companies and cannot be applied directly to Türkiye, but it offers a useful benchmark for how agent projects mature. Alongside the rise in agent use, the results give weight to controls such as cost reviews, token budgets and defined areas where agents may not decide without human approval. For companies in Türkiye planning agent pilots, this suggests it could pay to start by sorting out data access and to put in writing which decisions must always be approved by an employee.

For agent projects with data access and approval steps defined from the start, see our custom AI agents for companies page.

Sources

FU
Ferhat Uğurlu · YazSis Team

Builds enterprise software, AI and digital transformation products together with the YazSis team.