Skip to content

The AI You Bought Sits Idle. The AI You Didn't Buy Is Everywhere.#

SEO phrases: AI adoption, shadow AI, AI governance, change management, adoption-two-halves.

A shrink-wrapped, unused AI terminal on one side of a desk and a swarm of phones and laptops running AI chats on the other. Two halves, one strategy.

Last week I wrote down the five lines I'd want on paper before AI agents start acting inside a company's software. Approved tools, approved accounts, approved systems, an approval line, a log.

That page solves half the problem. The half where people are using AI and you don't know it.

The other half is the one I keep hearing about from owners. They bought the software. They ran the training. Nobody uses it.

Both halves live in the same building. AI adoption isn't a tooling problem. It's a people problem wearing a software costume.

The eager half goes around you#

Deloitte surveyed 25,000 UK workers in September. 63% have used AI tools at work. Almost one in ten have used a tool their employer banned or would frown on. And 63%, the same share that uses AI at all, worry their manager will decide AI can do their job (Deloitte, via Bloomberg and Claims Journal, September 16, 2026).

Read those two numbers together. The people using AI on the quiet aren't stealing time. They're afraid of what it says about them if they're seen doing it.

At larger companies the number is worse. In a June survey of 1,250 office workers at firms over $500M in revenue, two-thirds said they had used AI tools they believed were not permitted. 77% said their company's AI restrictions were limiting their career (PagerDuty and Wakefield Research, June 11, 2026).

These are your best people. The ones who solve their own problems. They tried a chatbot, tried a no-code app builder, tried whatever got the job done, because waiting for the policy meeting wasn't an option. They aren't waiting for permission. They're waiting for a better system than the one they built in secret.

You didn't lose them to a competitor. You lost them to a browser tab.

The hesitant half ignores what you bought#

Now the other side.

Gallup asked US employees in February. Half use AI in their role at least a few times a year. 13% use it daily. Only 41% say their organization has integrated AI at all, and among those that have, about one in ten employees strongly agree it has changed how work gets done (Gallup, February 2026).

The enterprise numbers say the same thing from the buyer's chair. Deloitte's survey of 3,235 leaders found sanctioned AI access grew from under 40% of workers to under 60% in a year. Among the workers who have access, fewer than 60% use it in their daily workflow, a pattern Deloitte says is largely unchanged from the year before (Deloitte, State of AI in the Enterprise 2026, fielded August to September 2025).

Even Microsoft runs into this. Fewer than 4.5% of the 450 million Microsoft 365 customers pay for Microsoft 365 Copilot (Fortune, May 29, 2026). That is not a knock on the tool. Microsoft put it inside software everyone already has. Presence still didn't turn into use.

That's the part owners get wrong. They think the purchase is the adoption.

It isn't. A license is a line item. If you can't say who owns the outcome, what data is allowed in, and where a person still says yes, you don't have an adoption plan. You have a subscription.

Adoption is when the receptionist, the estimator, and the controller all use it on a Tuesday without being told to.

It has a name#

Talking with my team this week, we kept landing on the same two words for the problem: AI adoption.

It means three different things depending on who's in the chair.

To the owner it means productivity. Same people, more output.

To the IT lead it means governed. Who used what, which model, on which data, with a record you can show an auditor or an insurer.

To the employee it means: I can log in and get help with my job, today, without setting anything up.

Most rollouts serve one of those three and starve the other two. IT locks it down and the eager half goes around it. The owner mandates it and the hesitant half nods and keeps working the old way. Leadership calls the result uneven. It isn't uneven. It's unmanaged.

Five moves that put the company on one page#

If I were rolling AI out to a 200-person company this quarter, this is the order.

1. One front door.#

Give everyone one sanctioned place to use AI, on company accounts, connected to the tools they already work in.

The eager half stops going around you the day the approved path is better than the workaround. The IT lead gets the log for free, because everything runs through the door.

2. Zero setup for the hesitant.#

The person who has never touched AI should log in and ask a question. No system prompt. No credentials to configure. No agent to design.

If the first screen asks them to build something, you've lost them. Simple and useful is the hard part. Anybody can ship simple.

3. Let the builders build, then share.#

The proficient half will write an app, a skill, a repeatable agent for the thing they do every week. Good. Give them the room.

Then make sharing the default. One person figures out the best way to handle a vendor invoice, and every person on the team has it the next morning. That is where the productivity number the owner wants actually comes from.

Same policy, same monitoring, two on-ramps.

4. Measure it like a rollout, not a purchase.#

Who's using it. Which model. What it costs. Which departments went quiet.

That dashboard is the governance IT asked for and the productivity proof the owner asked for. Same report, two readers.

For the hesitant half, make human review the feature, not the apology. Anywhere money, customers, or records are involved, the agent prepares the work and a person gives the yes. That is the line that lets a careful employee trust the tool.

5. Start with the work, not the data lake.#

Last week I sat in a room of operators worried about organizing years of data before AI could help them.

Skip it. Put AI on the front end of what people do every day and capture the workflow as it happens. What your team does this week tells you more than what's buried in a decade of old files.

How we run it#

We run JOV on this. Every person on our team logs into the same place. The agents draft, a person approves. This post was drafted by our marketing agent, from a transcript of a meeting I sat in, and I'm the one who read it and hit publish.

We built a governance policy with a foundation's board from scratch. The five lines from last week came out of that work. These five moves are what comes after the policy exists and the company still has to use the thing.

If half your team is racing ahead and half is digging in, you don't need another tool pitch. You need one playbook both sides can live with.

Let's map yours#

If you want to know which half of your company is where, schedule a 30-minute discussion. We'll list the AI tools already in use, find your eager and hesitant groups, and design the first front door: one place, company accounts, connected to the systems your people already work in.

Sources#