You Bought the AI. Nobody Changed the Work.

Nearly everyone has AI, and almost nobody changed the work

There is a popular way to describe AI adoption in 2026: it is like paying for a gym membership and never going. Companies buy licenses, add a few tools a vendor recommended, and then wonder why the quarterly numbers look the same. The analogy lands because it matches what the research shows.

Start with the first half of the claim, that around 90% of companies use AI. It holds up, with an important qualifier. McKinsey's State of AI in 2026, based on 1,719 respondents in 97 countries, found "nearly nine in ten respondents report regular use of AI in at least one business function." The U.S. Census Bureau tells a different story: 19.8% of U.S. firms used AI in the two weeks before its May 2026 survey, rising to 37% among firms with 250 or more employees.

Sources for the table: McKinsey, NBER, the Federal Reserve for the Atlanta Fed estimate, and the Census Bureau.

Both are right. A Federal Reserve note from April 2026 explains that the biggest source of the gap is "the survey goal, unit of analysis, and corresponding sample distribution." Executive surveys mostly reflect big companies; the Census counts every firm, and most firms are small. If you run a large or mid-size business, assume your competitors have AI. The question is whether they are using it differently than you.

The gap between having AI and getting value from it

The second half of the claim, that most companies have not built AI into how they work, is where the data gets sharp.

McKinsey found that 80% of respondents say AI has improved their individual productivity, but only 37% attribute any EBIT impact to it, a figure it describes as "essentially unchanged" from last year. High performers, companies getting real enterprise value, are 6% of respondents. An NBER study of nearly 6,000 executives in the U.S., U.K., Germany and Australia found 69% of firms actively use AI, yet nine in ten executives reported no impact on their firm's employment or productivity over the past three years.

Deloitte's State of AI in the Enterprise 2026, a survey of 3,235 business and IT leaders, put the problem in one sentence: 37% "report only using AI at a surface level with little or no change to underlying business processes," while only 30% are redesigning key processes around AI. And only 25% have moved 40% or more of their AI pilots into production.

Marketing is no exception. Gartner's 2026 CMO Spend Survey found CMOs now allocate an average of 15.3% of marketing budgets to AI initiatives, yet 70% say their internal marketing processes are not mature enough to implement and scale it.

The risk is that CMOs invest in AI tools faster than they build the data foundations, processes, governance and talent required to scale them.

That was Ewan McIntyre, VP Analyst at Gartner. We looked at where that hidden work ends up in why AI isn't working at your company. This article picks up the next question: what the companies getting results do differently.

What the companies getting results do differently

The pattern is consistent across every major 2026 study. The winners change the work, not just the toolset.

McKinsey reports that nearly three-quarters of its high performers have fundamentally redesigned workflows because of AI, up from 55% a year earlier, compared with about a quarter of other companies. Its description of the difference is worth reading twice.

High performers redesign workflows that are enabled by AI rather than insert AI into existing ones.

BCG reached the same conclusion from the worker side. Its 2026 AI at Work study of 11,749 workers in 14 markets found companies pursuing workflow redesign were 24 percentage points more likely to see measurable business improvement. A clear strategy lifted measurable impact by 25 points; better tools alone moved it about 5.

Microsoft's 2026 Work Trend Index, a survey of 20,000 AI users, found organizational factors such as culture, manager support and talent practices carried about twice the predictive weight of individual factors in reported AI impact. Microsoft is careful to call that an association, not proof of cause. Its section for leaders opens with a line that sums up the whole argument: "The job of every leader is to rearchitect work."

Two Wharton and University of Washington professors diagnosed the failure mode in Harvard Business Review in September 2026: companies "automate individual tasks rather than redesign the workflows that produce business value." Automating a task inside a broken process gets you a faster broken process.

The five stages companies move through

A useful way to see where you stand is as a progression through five stages. No single study measures this ladder, so the figures below come from different surveys with different samples. Treat them as signposts, not a precise funnel.

Sources for the table: McKinsey, Deloitte, Gartner's Hype Cycle for Agentic AI and Gartner's marketing automation survey.

Most companies are crowded into the first two stages. Gartner says agentic AI now "sits at the Peak of Inflated Expectations," with only 17% of organizations having deployed AI agents and more than 60% expecting to within two years. In marketing, leaders told Gartner they expect AI automation of marketing work to more than double, from 16% in 2026 to 36% by 2028.

The jump that matters is not from tools to agents. It is from adding AI to existing steps to rebuilding the process, which can happen at any stage. A team that redesigns one workflow around a simple assistant will usually outperform a team that bolts five agents onto an unchanged process.

About those 10x gains

The popular version of this argument says the companies redesigning workflows see 10x gains. That number does not survive contact with the research, and repeating it sets teams up to feel like failures when they get a solid 20%.

The best-known field study, of 5,179 customer support agents and published in the Quarterly Journal of Economics, found an assistant raised productivity by 14% on average and 34% for novice workers. The executives in the 2026 NBER study expect AI to raise their own firm's productivity by an average of 1.4% over the next three years.

Those are meaningful numbers. A 14% gain across a support team is worth a serious redesign effort. They are just not 10x, and the honest framing is that redesign turns modest, scattered individual gains into measurable business results.

Ask which workflow should disappear

Most companies start with the wrong question: which AI tool should we buy? The better question is which workflow should disappear, or at least which steps, handoffs and approvals no longer need to exist once AI can do the first draft, the research or the routing.

Here is a four-step way to rebuild one workflow, drawn from the research above.

Step one: pick one workflow and map it honestly

Choose a workflow that is frequent, measurable and painful, such as weekly reporting, content briefs or lead routing. Write down every step, every handoff between people, and how long each one takes today. Pick one and commit, rather than trialing ten tools in parallel; you need a clean before-and-after to know whether the redesign worked.

Step two: assign each task to whoever does it best

The instinct is to put a human and an AI on every task together. The evidence says that is often the worst option. A meta-analysis of 106 experiments in Nature Human Behaviour found human and AI combinations performed significantly worse on average than the best of humans or AI alone, with losses on decision tasks and a more positive pattern on creation tasks. Decide, task by task, whether AI, a person, or a combination does it best, and design accordingly.

Step three: set the human's role by the stakes

The HBR authors suggest choosing among four modes, which they call Assist, Approve, Audit and Automate, based on "the stakes and reversibility of potential errors," and designing for the "unhappy path," meaning how the system escalates exceptions and reverses mistakes. Anthropic's research on how people supervise its agents points the same way: "effective oversight doesn't require approving every action but being in a position to intervene when it matters." Low-stakes, reversible work can run with spot checks; anything client-facing or hard to undo gets a named approver.

Step four: retrain the team on the new process

Giving everyone access and hoping they figure it out is the gym membership problem in its purest form. Gartner found 80% of CMOs say staff fear and anxiety is a barrier to AI experimentation. Microsoft found 45% of AI users feel it is safer to focus on current goals than to redesign work with AI, and only 13% are rewarded for reinventing work even when results fall short. Train people on the new process step by step, have managers use it visibly, and reward the attempt as well as the result. Our guide to better AI prompts and saved context is a practical starting point for the training itself.

Where the popular advice goes too far

The core argument is right. A few parts of the popular version are worth pushing back on before you act on them.

We have written before about what happened when companies tried to replace whole teams with AI in should AI replace your SEO team. The short answer from those cases matches the data here: redesign the work with people in it, then shrink the parts that no longer need them.

Where to start

Your AI subscriptions are not the asset. The workflows you rebuild around them are. The companies pulling ahead in 2026 are not using better tools than everyone else; they are the ones willing to change how the work gets done.