McKinsey 2026
Agent Scaling Surges, EBIT Impact Stays Flat

McKinsey's annual State of AI survey, published August 25, confirms a pattern that has frustrated boardrooms for two straight years: record AI spending and rising personal productivity, yet the share of companies that can point to any bottom-line impact has barely moved. This year adds a sharper tension — agent scaling is accelerating at large enterprises while attributable earnings impact stays flat.

The data: scaling is up, EBIT is not

The survey, released by McKinsey's QuantumBlack division, covers 1,719 respondents across 97 nations (fielded May 4 to June 8, 2026). Two numbers together tell the story: among respondents at firms with revenue above $1 billion, 40 percent now scale agents in at least one business function, up from 27 percent last year; yet only 37 percent attribute any earnings (EBIT) impact to AI — essentially unchanged. Under a stricter definition, just 6 percent of organizations qualify as AI high performers — attributing at least 5 percent of EBIT to AI and calling the impact significant — while 63 percent report no measurable enterprise earnings impact at all.

The disconnect shows up most plainly at the function level: roughly 90 percent of specific function-level AI use cases remain stuck in pilot rather than scaled to production. Cost reductions cluster in supply chain, service operations, and manufacturing — bounded jobs with measurable outputs. Revenue gains show up in marketing, sales, and product development. Everything else — the broad enterprise chatbot rollouts most companies can point to — is still producing individual satisfaction without organizational impact.

Personal productivity is not corporate earnings

Eighty percent of respondents who use AI in their role say it improved their personal productivity, but that is not aggregating into the EBIT gains CFOs and investors track. McKinsey's explanation is structural: horizontal tools — chatbots, copilots, writing assistants — are easy to scale and genuinely help individuals, but the time they save stays at the worker's desk rather than flowing to the income statement. Enterprise-level earnings impact requires redesigning workflows around AI, not adding AI on top of existing workflows. Most organizations have done the latter.

McKinsey's own deployment is a live demonstration: CEO Bob Sternfels disclosed roughly 25,000 AI agents running alongside 40,000 human consultants, with a target of near 1:1 parity by year-end. Back-office functions shrank 25 percent while client-facing roles grew 25 percent — and back-office output rose 10 percent with fewer people. Personal efficiency did not simply sum into organizational results; the processes themselves were reworked.

The token multiplier: a new operating constraint

Scaling brings a new financial constraint: the cost structure is shifting from software licensing to inference-token consumption. Nearly a third of respondents said their organizations opted for in-house AI builds rather than buying at least one software product, moving capital from SaaS license fees to inference fees — a less predictable, more variable cost. One in five said AI-related operating costs, including tokens, have already constrained their use of the technology.

The reason is architectural. Gartner's March 2026 analysis of agentic AI cost dynamics estimates that a workflow which reasons, calls external tools, verifies outputs, and self-corrects generates 5 to 30 times more inference tokens per task than a standard chatbot answering the same query. At enterprise scale, that multiplier turns manageable AI budgets into significant line items. CIOs are beginning to measure the full cost of completing a workflow rather than seats or token growth — and asking which tasks justify frontier-model pricing.

The high-performer divide: workflow redesign

The survey points to one distinguishing practice above all others: nearly three-quarters of high performers redesigned workflows around AI, versus about a quarter of all other respondents. Identify a specific business process — a customer escalation workflow, a claims review pipeline, a supply-chain exception routine — where AI owns multiple steps end to end rather than assisting humans at individual steps. Pair that with a measurable baseline established before deployment, visible senior leadership ownership, and a willingness to reallocate headcount rather than just adding AI capacity on top of existing operations.

For CIOs and CFOs, this survey's value is separating two facts: agent scaling is real, accelerating, and happening mostly in large firms; attributable EBIT impact is stagnant. The bridge between them — and the divide between high performers and the other 94 percent — is not a better model. It is workflow redesign.

References

• TechTimes: Record AI Spending Can't Move Earnings Needle for 94% of Enterprises, McKinsey Finds (2026-08-26) — https://www.techtimes.com/articles/325590/20260826/record-ai-spending-cant-move-earnings-needle-94-enterprises-mckinsey-finds.htm
• Quasa: AI-Agent Scaling Rises as Reported EBIT Impact Stays Flat (2026-08-29) — https://quasa.io/insights/ai-agents-scale-at-40-of-large-firms-but-profit-impact-is-flat
• Gartner (2026-03): agentic AI cost dynamics, 5 to 30x token multiplier per task (as cited by TechTimes)