The headline numbers in McKinsey's 2026 State of AI survey of 1,719 leaders are easy to summarise and hard to reconcile. Eighty-eight percent of organisations use AI in at least one function. Eight in ten respondents say it has improved their own productivity. And the share attributing any contribution to EBIT is 37 percent, unchanged from a year ago. Only six percent qualify as high performers, meaning five percent or more of EBIT attributed to AI. Conviction, as the report puts it, is growing faster than the returns anyone can point to.
1. Why the two numbers diverge
Individual productivity does not add up to firm profit unless something else changes. A lawyer who drafts faster still bills the matter; an accountant who closes faster still has the same fee. Time saved is converted into money only when the firm redesigns the work: reprices the engagement, reallocates the people, or takes on volume it could not take before. McKinsey's high performers are roughly three times as likely to have fundamentally redesigned workflows. The rest have added a tool to an unchanged process and measured the process.
2. Agents are scaling, unevenly
The share of large organisations scaling agents in at least one function rose from 27 to 40 percent in a year. Among smaller organisations it stayed flat at 22 percent. Agents are most often scaled in IT, knowledge management and software engineering, the functions with clean data and clear handoffs. Client-facing professional work, with its messy inputs and judgement calls, is later in the queue, which is exactly why the firms doing it now are the ones to watch.
Productivity is a personal number. The P&L is an organisational one. Only redesign connects them.
3. Cost has become a constraint
For about one in five organisations AI-related operating costs are now limiting use. This is new. Two years ago the constraint was capability; today a firm can run out of budget for inference before it runs out of ideas. Model routing, effort settings and hard spend caps have moved from engineering niceties to line items a COO reads.
4. The workforce number that did not happen
Last year 32 percent of respondents expected AI to shrink their total workforce over the following twelve months. This year 14 percent report that it did. The gap tells the same story from the other side: firms have not restructured, so the headcount effect has been smaller than feared, and the financial effect smaller than hoped.
5. For firm leadership
- Measure at the engagement, not the individual. Hours per matter, write-offs per fixed fee, days to close. If those have not moved, the productivity gains are being absorbed.
- Pick one workflow and redesign it end to end, including price and staffing, rather than distributing tools evenly across the firm.
- Put a number on inference. If nobody can say what the firm spent on AI compute last month, cost will surprise you in the second half of the year.
- Be one of the 22 percent that moves. The survey says mid-sized organisations have stalled on agents. That is a competitive opening for the ones that do not.
Hive Newsroom follows what is changing in AI and professional services. Sources are linked in the text; figures are as published at the time of writing. Comments and corrections: press@get-hive.ai. Back to the .