McKinsey State of AI 2026: What It Means for You

AI is helping people work faster, but company profits are moving more slowly. McKinsey’s August 2026 survey finds that 80% of people using AI at work feel more productive. Only 37% of respondents attribute positive company-wide earnings impact to AI. About 6% qualify as high performers.

Company-wide results

  • 37% of respondents attribute positive company-wide earnings impact to AI
  • About 6% qualify as high performers

Workforce: expected and actual

  • 32% of respondents at AI-using organizations expected a workforce decline
  • 14% say AI contributed to an overall workforce decline over the past year
Painted illustration of fictional tailoring workshop colleagues planning their work together in Bahrain
Painted illustration of a fictional Bahrain workshop. It does not depict a surveyed business.

For an owner, our reading is simple: a faster task is a useful start, but the job is not finished there. Look at what happens after that task. Does the customer get an answer sooner? Does someone still have to redo the work? That is where we would look before buying another tool.

Who did McKinsey ask?

The McKinsey State of AI report, On the road to ROI, is dated August 2026. McKinsey surveyed 1,719 participants in 97 nations from 4 May to 8 June 2026. It weights the answers by each respondent’s nation’s contribution to global GDP.

Read the findings as people’s accounts of their workplaces, rather than a count of every business in the world. The questions also have different bases. McKinsey’s personal productivity question goes to people who use AI in their roles. Its company-wide scaling chart covers organizations using AI. Keeping those groups separate helps explain what the results can tell you.

Keep the earnings result beside the productivity headline. Feeling faster and earning more are different outcomes. EBIT means earnings before interest and taxes. A member of staff can finish a draft quickly while the business still waits for an approval, a correction or a customer’s reply.

What changed since 2025?

The McKinsey comparison shows wider deployment alongside little movement in earnings impact. In McKinsey’s chart of organizations using AI, the scaling share moves from 38% in 2025 to 44% in 2026. McKinsey describes positive EBIT impact as essentially unchanged, at 37% in 2026.

The table keeps each measure attached to its question. Read the earnings row as McKinsey’s assessment of the year-on-year change.

On a phone, swipe the table sideways to see both years.

McKinsey measure20252026PDF page
At least scaling AI, among organizations using AI38%44%5
Respondents attributing positive company-wide EBIT impact to AI“essentially unchanged from 2025”37%12
AI high performers, share of all respondents“remained flat at about 6 percent”About 6%3, 17
Respondents at AI-using organizations expecting an overall workforce decline over the following year32%39%25
McKinsey respondents reporting an AI-linked decline in workforce size, against the prior expectation32% expected workforce reductions over the following year14% report an AI-linked decline over the past year25

Source: McKinsey, The state of AI in 2026. Page numbers are the report’s printed pages. Each workforce expectation concerns the year following that survey.

Ask about the result of wider use. If more people have access to AI, ask what has improved for the business. Access, use and financial benefit deserve separate answers. A busy team may value less repetitive work even before it can point to a change in earnings.

Are big companies pulling ahead on agents?

McKinsey’s larger-company group has annual revenue of at least $1 billion. McKinsey’s larger organizations have agent scaling in at least 1 function at 27% in 2025 and 40% in 2026. McKinsey describes the smaller-company result as essentially flat, at 22%. Its agent chart rebases the answers to the full sample.

“Smaller” here does not mean a corner shop. It is the group below McKinsey’s revenue threshold. Borrow a useful working habit from the comparison, rather than treat a large-company deployment rate as your own target.

There is also a change in buying decisions. Among respondents at organizations using AI, McKinsey finds 32% say their organization skipped a software product or feature purchase because it could build the function with AI coding tools.

McKinsey also finds about 20% of respondents at AI-using organizations say operating costs constrained AI use. Our suggestion is to include the work after the first working version in any build-or-buy decision. Ask who will fix failures, handle access and keep the software running. The NCSC’s AI security guidance covers operation and maintenance as well as development. That makes it useful reading before committing to an in-house tool.

What do the high performers do?

The hand-offs

Add a drafting tool

  • The old approval queue may remain

Redesign the workflow

  • Decide who reviews the draft and where the finished work goes

McKinsey counts a high performer as a respondent attributing at least 5% of company EBIT to AI and describing its impact as significant. This is a defined survey group, rather than a label for anyone who uses a lot of AI.

McKinsey finds nearly three-quarters of high performers redesign workflows because of AI, up from 55% the previous year. Among other respondents, McKinsey puts workflow redesign at one-quarter. McKinsey describes high performers as changing workflows rather than layering AI onto existing work.

Our reading is that the hand-offs deserve as much attention as the tool. If you add a drafting tool, the old approval queue may remain. If you redesign the workflow, you can decide who reviews the draft and where the finished work goes. The survey shows an association, not a promise of the same return.

We would start by drawing the job from the customer’s request to its completion. Mark where the work waits. Then decide which part AI should handle and which decision belongs to a person. Choose an outcome that matters to the customer before choosing a product.

Did the expected job cuts happen?

McKinsey’s workforce comparison separates a forecast from an outcome. In its prior survey, 32% of respondents at AI-using organizations expected a workforce decline. In the new survey, McKinsey finds 14% say AI contributed to an overall workforce decline over the past year.

Looking ahead, McKinsey finds 39% of respondents at AI-using organizations expect an overall decline over the next year. McKinsey finds 43% expect little or no change. Separately, McKinsey says 13% express anxiety about their own career prospects.

For a worker, our advice is to keep the forecast in perspective while learning how the work is changing. Ask your manager which tasks need better results and what good use of AI would look like. For an owner, discuss the change in the work with the people doing it before treating a forecast as a staffing plan.

What does it mean for a small business?

The U.S. Census Bureau offers a different view of business AI use. Its data from December 2025 to May 2026 put overall use between 17% and 20%. Census says fewer than 20% of firms with 4 or fewer employees used AI. These are Census figures, not McKinsey findings.

The Census question asks about AI use in the past 2 weeks, in any business function. Its coverage and question differ from McKinsey’s global workplace survey, so the percentages should not be used as competing estimates of the same thing.

The OECD’s research on small businesses adds another useful perspective. It identifies fit with the work, information concerns and staff skills among barriers to generative AI use. “Should we use AI?” is too broad a starting question. Ask where it could help your particular team.

McKinsey’s finding on workflow redesign offers a useful way to think about a small-business trial. Nearly three-quarters of its high performers redesign workflows, compared with one-quarter of other respondents. The example below puts that distinction into a familiar job.

Our example (illustrative): A tailoring workshop quote. Imagine the fictional workshop in the banner receiving a request for matching staff uniforms. All times below are made up to show the hand-off, not results from a real business.

What changes when the owner changes the hand-off?

Compare the same job before AI, with an AI draft, and after the team redesigns approval. The owner still decides whether the quote can go to the customer.

Our example and our calculation: time spent or waiting, in minutes
StepWhoBefore AIWith an AI draftAfter redesign
Record the requestReceptionist555
List fabric and workTailor101010
Prepare and review the draftAssistant2055
Wait for approvalOwner12012015
Our calculation: total minutesWhole process15514035

After redesign, the receptionist alerts the owner when a complete draft is ready. The owner reviews it within the shorter approval wait shown above. The assistant’s drafting time includes review of the AI draft.

The AI draft makes preparation faster, but leaves the long approval wait in place. Changing the hand-off addresses that wait. The owner’s approval remains part of the process.

Track 1 measure in the trial: elapsed time from a complete job brief to an approved quote. Record those timestamps for the old and new process, including any corrections. That shows whether faster drafting turns into a faster answer for the customer.

Agree what a person must review before work reaches a customer. For another setting where those boundaries matter, see our guide to AI for teachers. The NIST AI Risk Management Framework offers voluntary guidance for managing AI risk.

Common questions: what should you take from the report?

Does the report say AI is failing?

No. McKinsey finds personal productivity benefits alongside a much smaller group attributing company-wide earnings impact to AI. They measure different outcomes.

Should a small firm copy a high performer?

Borrow the focus on workflow design and measure your own result. The survey does not guarantee that a practice will produce the same return in your business.

Is an expected job cut an actual job cut?

No. McKinsey asks about past changes and future expectations separately. Keep the time period with the answer whenever you quote the finding.

Sources

Published

Chat on WhatsApp