Two exhibits from the same survey
McKinsey's State of AI 2026 published on 25 August, based on 1,719 participants across 97 countries fielded in May and June. Two of its exhibits describe the same AI rollout the way people at either end of the org chart lived it, and they land on opposite conclusions.
Executives and senior managers: 81% say AI has improved their productivity, 31% report AI-related strain.
Mid-level managers and individual contributors: 54% and 47%.
The gain rises as you go up the chart. The strain rises as you go down. One technology. One survey. Neither exhibit is a mistake, and neither group is wrong about their own experience of the tools.
What the second exhibit is, and what it is not
The 47% is not a temperature check. The survey did not ask that group how they felt about AI; it asked what it had done to their work. The finding is that the majority end of the chart is reporting added workload, career worry and stress — not offering an opinion on the technology. The gap between the two exhibits is a description of what the rollout did to the workforce, told by the workforce.
A rollout that took its own reading and treated it as the whole reading
The people who chose which tools to buy, which workflows to change and how fast to move are the people the top row describes. The people who were told which tools to use, what changed in the workflow and how fast the pace now runs are the people the bottom row describes. Both readings are accurate.
If the leadership team is running a program shaped by their own experience of the tools — 81 and 31 — they are designing for the 20% of the workforce whose numbers match theirs. That is not something the next all-hands can fix by explaining the tools better. It is a design gap: the same rollout reads as leverage from a director's chair and workload from the chair of the person doing the work, and misreading the second reading is what will decide whether the productivity claim eventually shows up on a P&L or does not.
What a rollout designed for the other end looks like
The organizations that will bring the 54% closer to the 81% are the ones whose rollout is redesigned around the workflow of the person doing the work, not the person choosing the tool. In practice that means fewer new tools introduced faster, and more time spent on how one already-adopted tool fits inside the work an individual contributor is being measured on. Fewer platform demos; more time with one person, on one task, watching where the tool actually helps and where it adds a review step that was not there before.
It is a smaller and cheaper program than most of what is currently being tried, and it moves the number that matters.
Redesign for the person doing the work, not the person choosing the tool: Design the rollout for the people it lands hardest on →
Keep reading: The ones who kept score · Who did most of the thinking
This is the thinking behind a Trusted AI Culture.
Sources: McKinsey QuantumBlack, "The State of AI in 2026: On the road to ROI" (25 August 2026), n=1,719 participants across 97 nations, fielded May 4 to June 8 2026, weighted by contribution to global GDP. Productivity is self-reported ("Improves my productivity"); the strain figure aggregates respondents reporting one or more of stress, workload, or career anxiety.
Written with AI in the loop: my idea, AI drafted and sharpened, my judgment on the way out. Every word is mine to stand behind. — Darren