Here's a question I hear from executives almost every week: are we behind on AI?
You're not alone in asking. In the American Psychological Association's 2026 Work in America survey, 37% of working adults said they worry about falling behind their coworkers if they don't use AI. (APA, September 2026)
That number measures worry. It says nothing about who's actually ahead. And when I look at this year's research, "behind" turns out to be the wrong question for most companies. The better one is: behind on what?
Four numbers worth knowing
81% vs 25%
Leaders at companies already using AI who say it sped up decisions (more than half of them call the gain slight), vs. those using it to deeply transform how the business works. (Deloitte, Oct 2026)
87% vs 47%
Senior decision-makers whose companies encourage using AI agents, vs. those with clear governance and controls for them. (OneTrust, Sep 2026)
30% / 15%
U.S. employees who use AI at work a few times a week or more / every day. (Gallup, Jul 2026)
78% vs 44%
Frequent AI use among employees who strongly agree their manager actively supports it, vs. those who don't. (Gallup, Apr 2026)
These come from different surveys of different people, so don't stack them into one funnel. Read together, they say the same thing. Access is spreading fast. Turning access into changed work is moving much slower.
A license is not a strategy
The most common AI rollout I see goes like this. The company buys seats for ChatGPT, Copilot or Claude, sends a kickoff email, maybe runs a lunch-and-learn, and counts the licenses as progress.
A few months later, a small group is using it to produce real work, and most people are using it like a better search box. Leadership can't tell whether the company is ahead or behind, because nobody defined what the tool was supposed to change.
Gallup's 78% vs 44% is the clue. Inside companies that already provide AI, whether people use it regularly tracks closely with whether their manager is actively in it with them. (Gallup is clear that's a correlation, not proof of cause.) Seats don't spread a practice. People do.
Three things that changed this year
1. Agents need a supervisor, not just permission. Agents are now doing real work inside companies. In Dynatrace's late-2025 survey of 919 enterprise leaders running them, 69% of agentic AI decisions were still verified by a human. (Dynatrace, Jan 2026) And a 2026 study of 15 models found that as they got more capable, they got only a little more reliable. (Rabanser et al., 2026) Finishing a task once is different from doing it right every time. Before an agent runs a piece of work, someone should write down what it may read, draft, change and send, who reviews it, and when it stops.
2. Build vs. buy just moved. In McKinsey's 2026 survey, 32% of respondents said their company had decided against buying at least one software product because they could build it themselves with AI coding tools. (McKinsey, Aug 2026) As a CTO, I'll add the part that doesn't show up in the demo: building got cheap, but owning didn't. Someone still maintains it, secures it and answers for it when it breaks. Name that person before you build.
3. The entry-level ladder is getting shorter. Stanford's Digital Economy Lab, using payroll data, found employment for 22-to-25-year-olds in the most AI-exposed jobs running about 19% behind a less-exposed comparison group, mostly through slower hiring. (Stanford, Aug 2026) That isn't a layoff rate, and the researchers don't claim AI caused it. But it raises a question every leader should answer now: if AI writes the first draft, how does your next generation learn the judgment to check it?
What "keeping up" actually looks like
You don't keep up by tracking every model release. You keep up by getting one piece of work right, then doing it again.
Pick one recurring task. Something that happens every week and has an owner. A weekly status report, a client recap, a prospect brief.
Write down the starting point. How long it takes today, what "good" looks like, and who checks it.
Decide what AI should do with it. There are five honest answers: leave it alone, redesign it, have AI assist a person, delegate it to AI with review, or automate it. Leaving it alone is a valid answer.
Set the boundaries. What AI may read, draft, change and send, and what needs a human's OK before it leaves the building.
Run it for a month, then decide. Track quality and rework, not just speed. Then keep it, fix it, expand it or stop it, and teach it to a second person.
That last step is where most companies drop the ball. A practice that lives in one person's head isn't capability. It's a dependency.
There's good evidence that a focused rollout like this pays off. In a study of 5,172 customer support agents at one company, AI assistance raised issues resolved per hour by about 15%. (Brynjolfsson, Li & Raymond, QJE, 2025) That was one job, at one company, measured carefully. That's the scale where results show up.
So, are you behind?
Probably not on access. Most companies reading this already bought the tools. The gap worth worrying about is whether you can take one piece of work, change it with AI, prove it helped, and repeat that across the company.
That's what RE+SHAPE is for. It's a seven-step way to take one recurring piece of work and decide whether AI should touch it at all, and if so, exactly how.
Not sure where to start? The workflow finder gives you the top three for your role in about 8 minutes.
Keith