AI Agents Scale at 40% of Large Firms—but Profit Impact Is Flat

Large enterprises are scaling AI agents faster, but the reported financial payoff has not kept pace. McKinsey’s global survey published August 25 found that 40% of respondents at organizations with more than $1 billion in annual revenue were scaling agents in at least one business function, up from 27% a year earlier, while 37% of all respondents attributed at least some positive EBIT impact to AI—essentially unchanged from 2025.

The gap is clearest when company-level results are compared with employees’ assessments of their own work. ITPro’s August 26 analysis highlighted the contrast between the 80% who said AI improved their individual productivity and the unchanged 37% share attributing an earnings contribution to the technology.

What the 40% scaling figure measures

The headline figure is a survey response, not an independently verified count of production systems. It covers people at large organizations who described agents as being in the scaling phase within one or more functions; it does not establish that 40% of all billion-dollar companies have completed enterprise-wide deployments.

Scaling within a function is also narrower than scaling across an entire organization. An agent can move beyond experimentation in software engineering, IT or knowledge management while remaining absent from most other operations. The result therefore indicates a wider operating footprint, but not the volume of work completed, the reliability of the output or the economic value produced.

Company size matters to the adoption result. ETNews’ August 27 account of the survey placed the rise from 27% to 40% at large organizations beside an unchanged 22% scaling rate at smaller ones and noted that only 6% of all respondents qualified as AI high performers.

That split suggests the headline increase is concentrated rather than universal. Larger enterprises can have more workflows, capital and technical capacity available for deployment, but the survey does not isolate which of those factors accounts for the difference.

Personal productivity is not an EBIT calculation

The individual-productivity result records respondents’ judgments about their own performance. It may reflect faster drafting, coding, research or analysis, but it does not calculate whether saved time became additional output, lower staffing costs, greater revenue or a higher operating margin.

The EBIT measure asks about impact at the organizational level and is likewise based on attribution by respondents rather than audited financial statements. Software subscriptions, computing, integration, oversight and process redesign can absorb part of any task-level benefit, while time saved by an employee may simply create capacity for other work without reducing expenditure.

The high-performer category sets a more demanding standard: respondents must attribute at least 5% of EBIT to AI and describe the value created as significant. Its unchanged 6% share does not prove that the remaining organizations received no value; it shows that few respondents met both parts of that definition.

Deployment, perceived productivity and financial impact are therefore three separate measures. The survey does not provide a causal test matching scaled-agent deployments against audited earnings, so it cannot establish that agents produced—or failed to produce—a specific profit outcome.

Expected workforce reductions still exceed reported outcomes

The same distinction between plans and realized results applies to employment. TechTimes’ August 26 summary of the workforce findings said 14% of respondents at AI-using organizations attributed an overall workforce decline to AI during the preceding year, compared with the 32% who had expected reductions a year earlier; 39% now expect declines in the coming year, while 43% expect little or no change.

The comparison sets expectations in one annual survey against retrospective assessments in the next. It is not a matched audit of payroll records, and it measures whether respondents believed AI contributed to a change rather than whether employment moved for other reasons.

Agent scaling consequently cannot serve as a direct proxy for job losses. A company can expand an automated workflow while total employment remains stable because demand grows, responsibilities shift or human review remains necessary.

The unresolved question is conversion into financial results

The 2026 findings support a clear but limited conclusion: respondents at large enterprises describe substantially more agent scaling, and most respondents perceive individual productivity gains, yet the share assigning any positive EBIT contribution to AI has not increased. Those findings can coexist because they concern different subjects, scopes and standards of evidence.

The survey was conducted online from May 4 to June 8 among 1,719 participants in 97 countries, with responses weighted by each nation’s contribution to global GDP. That breadth makes it a global indicator of executive and employee perceptions, but it does not turn those perceptions into independently audited corporate performance.

What remains unknown is how much task-level benefit will eventually convert into sustained revenue, lower operating costs or wider margins. Later surveys can show whether respondents’ EBIT attribution changes, while company disclosures may provide firmer evidence where businesses separately quantify AI’s effects.

For now, the 40% result is evidence of survey-reported scaling among organizations above the revenue threshold—not proof of profit at scale. The flat 37% EBIT measure remains the counterweight: deployment is advancing faster than the financial impact respondents can currently assign to it.

Also read:

• AI Agents Ran 26 Minutes per Session—but That Is Not a Productivity Guarantee
• Goldman’s Claude Agents Moved Toward Launch, but Results Stay Undisclosed
• Genesis Mission Picks 278 Projects—but Funding Is Not Yet Final