Forty percent of large enterprises now say they are scaling AI agents across at least one business function, up from just 27 percent a year ago. That is a real jump. Yet the share of companies reporting any bottom-line impact from AI has barely moved, stuck at 37 percent for two years running. This is the AI agent scaling gap 2026, and it is the defining story to come out of McKinsey’s newly released State of AI survey.

For anyone building with or investing in autonomous AI systems, this gap is not a footnote. It is the central question of the year: why is deployment sprinting ahead while financial returns crawl behind it. In this article, we break down what McKinsey’s 1,719 respondents actually reported, where AI agents are delivering value today, why most organizations are not seeing it show up on the income statement, and what the small group of high performers is doing differently.

Enterprise AI Agent Adoption Is Accelerating Fast

The headline number is hard to ignore. Large organizations, those with more than a billion dollars in annual revenue, went from 27 percent scaling AI agents in 2025 to 40 percent in 2026. Smaller companies barely moved, holding flat at 22 percent. That widening split matters: enterprise AI agent adoption is increasingly a scale game, and the gap between big and small companies is growing rather than closing.

Chatbots remain the most widely scaled AI tool overall, with 47 percent of respondents saying their organizations have them running enterprise-wide. But agents and software coding agents are catching up quickly, each scaled by roughly two in ten respondents. Adoption is not evenly spread. Technology and media companies lead the pack, and within those industries agents are concentrated in IT, knowledge management, and software engineering. Consumer goods and retail companies lean on agents for marketing and sales, while advanced manufacturing firms deploy them in supply chain and production workflows.

This pattern lines up with what we have covered in our look at enterprise AI agent platforms: the companies moving fastest are the ones with existing digital infrastructure and technical talent to plug agents into. For everyone else, the on-ramp is still steep.

The AI Agent ROI Problem Behind the Adoption Numbers

Here is where the story turns. Despite the adoption surge, only 37 percent of respondents report any EBIT impact from AI use, a figure essentially unchanged since last year. Just 6 percent qualify as true AI high performers, meaning they attribute at least 5 percent of EBIT to AI and describe its impact as significant. That share has not budged either.

The AI agent ROI picture gets more interesting when you look at costs. About one in five respondents say rising operating costs, including token spend, are now constraining how much AI their organization uses. That is consistent across company sizes and industries, which suggests the token bill problem we detailed in our piece on AI agent operating costs is not going away as usage scales up. It may be getting sharper.

Individual employees, meanwhile, are seeing real gains. Eighty percent of respondents say AI has improved their personal productivity, and half say it helps them make better decisions. The disconnect is structural: personal productivity gains are not automatically converting into enterprise financial performance, and that conversion problem is exactly what separates the 6 percent of high performers from everyone else chasing the same technology.

What Closes the AI Agent Scaling Gap in Practice

McKinsey’s data points to a clear behavioral difference among high performers, and it has little to do with which model or vendor they chose. Nearly three-quarters of high performers report fundamentally redesigning their workflows around AI, compared with just a quarter of other companies. Most organizations are still bolting agents onto existing processes rather than rebuilding around them, and that shortcut appears to cap the returns available.

High performers also invest more aggressively and more broadly. They are more than three times as likely to be scaling agents across most business functions, and more than twice as likely to spend over 15 percent of their IT budget on AI. They manage risk more actively too, paying closer attention to unauthorized agent actions and technical vulnerabilities, a theme we explored in our earlier coverage of AI agent adoption and the hype cycle reality.

One trend worth watching closely: nearly a third of organizations, 32 percent, have now skipped buying a software product because agentic coding tools let them build the functionality in-house instead. Among high performers, that figure nearly doubles to 47 percent. If your business is still purchasing off-the-shelf tools by default, it is worth asking whether an internal agent could now do the job faster and cheaper, a question we dig into further in our build vs buy AI agents breakdown.

The Road Ahead: Confidence Is Outrunning Proof

There is a nuance in this data that deserves more attention than it usually gets. Workforce expectations are shifting upward again, with 39 percent of respondents now expecting AI-driven headcount declines in the coming year, versus 32 percent who expected the same a year earlier. But actual reported reductions over the past year came in at just 14 percent, less than half of what was predicted. Organizational conviction in AI agents is running well ahead of measurable outcomes, on both the cost-savings side and the headcount side.

That gap between expectation and evidence is not necessarily a red flag. It may simply reflect how long transformation takes relative to how fast confidence builds. But it is a reason for caution before assuming that scaling activity alone will translate into results on any predictable timeline.

Key Takeaways and What To Do Next

Three things stand out from this year’s data. First, enterprise AI agent adoption is genuinely accelerating, but almost entirely among large companies with the resources to redesign around it. Second, the AI agent ROI gap persists because most organizations are inserting agents into old workflows instead of rebuilding them, while rising token costs quietly cap the upside. Third, the minority of high performers who redesign workflows, invest more heavily, and manage risk proactively are pulling meaningfully ahead of everyone else.

If you are deploying or evaluating AI agents for your own organization, the lesson is not to slow down. It is to be deliberate about how you deploy. Explore more tools, frameworks, and coverage of the agentic AI landscape at BigAIAgent.tech to see how other businesses are closing this exact gap.

What would it take for your organization to move from experimenting with AI agents to actually seeing it on the balance sheet?

Source: McKinsey, “The state of AI in 2026: On the road to ROI,” August 25, 2026.