How to get employees to actually use AI

The licences are paid for. The operations team is still on spreadsheets, the logistics coordinators are still calling suppliers by phone, and the dashboard nobody opens is entering its second year. Usage reports get circulated, training sessions get rebooked, and the number does not move.

The usual explanation is that people are resistant to change, or that they need more training. Watch what actually happens when someone tries to use the tool. They get an answer, and then they need three approvals before they can act on it. The blocker is permission, not capability.

The systems around the tool are built — often without anyone deciding it — to prevent things from happening. Context sits behind management layers. Decisions are gated by title. Roles have tight boundaries. Approval cycles exist to distribute blame rather than to improve the outcome. The message every employee receives is that they are not trusted to act without supervision.

In that environment, an AI tool hands more capability to someone who still has to ask. You can give a person the most powerful analytical engine ever built and it changes nothing, because the thing they were short of was never analysis.

The Problem Isn't New — But AI Makes It Visible

Organizations have always limited information flow and decision authority. The logic was coherent: in a world of expensive, irreversible decisions, you centralize authority with the people who have the most context. Managers exist to relay information up and decisions down. The hierarchy is an information routing system.

AI breaks this logic. When tools can surface context instantly, the information asymmetry that justified centralized decision-making evaporates. The manager who added value by synthesizing reports now competes with a system that does it in seconds. The weekly strategy meeting where context was shared is redundant when the context is live and available to everyone.

But most organizations haven't updated the operating system to match. They've installed new tools on top of old power structures. The result: AI-generated insights flowing into decision processes still designed for a world where information was scarce and expensive to move.

What the Best Applied AI Company Learned

One company confronted this problem earlier than most — because their entire business model required deploying AI into organizations that didn't trust it. Government agencies. Defense contractors. Manufacturers. Banks. Environments where the default posture toward new technology is suspicion, and where adoption requires earning trust from operators who have every reason to resist.

Over 150+ enterprise deployments, they learned that culture and org design are necessary but not sufficient. Trusting your people is the right instinct. Flattening hierarchies helps. But these are two pieces of a five-part problem. The organizations that actually get AI into production — where it changes how work gets done, not just how reports get generated — invest in five capabilities simultaneously:

1. Strategy That Starts From Your Own Problems

Most AI initiatives are mimetic — launched because a competitor announced something similar, or an analyst recommended it, or a conference speaker made it sound inevitable. The problem being solved is “keep up” rather than “our specific operation has this specific friction point costing us this specific amount.”

The deployments that reach production start from someone embedding in the operational environment and identifying the actual problem — the one that makes Tuesday mornings miserable, not the one on the transformation roadmap. The discipline is ignoring what everyone else is building and asking what your operation actually needs.

2. Ownership Without Handoffs

Trust requires accountability. And accountability requires ownership — one person who carries the problem from diagnosis through production. Not a strategy team that defines the vision, a data team that builds the pipeline, an IT team that deploys it, and a change management team that handles adoption. Each handoff loses context. Each boundary creates a place where nobody is responsible for the outcome.

The structural answer is embedding a single person who sits with the operators, hears the stated problem, diagnoses the real one, builds the solution, deploys it, and stays until it works. When one person owns the full loop, the trust problem shrinks — because the operators know exactly who is accountable and that person knows exactly what success looks like.

3. Talent Built for Learning, Not Specialization

Autonomy only works if people can actually handle the problems they encounter. The typical enterprise AI team — domain specialist, ML engineer, data engineer, project manager — distributes the problem across four people, none of whom own the whole thing. Nobody has enough context to make a fast decision alone.

The alternative is selecting for technical generalists with exceptional learning agility — people who can walk into an unfamiliar domain (automotive manufacturing, hospital logistics, government procurement) and become dangerous within weeks. This is the talent profile that makes autonomy viable: someone who can earn operational trust fast enough to be given real problems, and solve them end-to-end without waiting for another team to handle their piece.

4. A Discipline for Navigating Resistance

Even with trust, autonomy, and capable people — deployments hit resistance. A plant manager feels their authority threatened. A compliance officer sees unmanaged risk. A VP quietly deprioritizes because the new system shifts budget control. The surface objection is always technical. The real resistance is about who controls decisions and whose expertise gets displaced.

Most “AI transformation” advice stops at “change the culture.” It doesn't tell you what to do on Thursday afternoon when the operations director crosses their arms and says “this won't work here.” The deployers who succeed have a trained discipline for reading what the organization is protecting — and for adjusting their approach in real time. These are specific, learnable skills with a theoretical foundation (drawn from improvisational theatre, of all places) that determine whether a working system gets adopted or becomes shelfware.

5. Principles That Survive Retransmission

The last piece is the hardest to scale: how do you maintain coherence across hundreds of people deploying into different industries, geographies, and organizational contexts? How do you ensure that the trust and autonomy you've built doesn't degrade into chaos as the organization grows?

The answer is encoding principles into mental models vivid enough to survive retransmission — distinctive metaphors and frameworks that compress operating logic into memorable form. Not values posters. Not acronyms. Language that lets a team member in Ohio and a team member in Virginia make consistent decisions without escalating. This is what keeps a high-trust, high-autonomy organization coherent at scale instead of drifting into fragmentation.

Beyond Culture and Org Design

The instinct is correct: AI transformation requires trust, autonomy, and agency. But these are outcomes of a system, not interventions you can announce at an all-hands. The organizations that actually achieve them have invested in all five capabilities simultaneously — strategy, ownership, talent, deployment discipline, and communication — because each one enables the others.

One company built a $250 billion market cap by solving these five problems together, across government agencies, hospitals, manufacturers, and financial institutions. The book is a complete decomposition of how they did it — and what any organization can extract from their playbook to get AI from announcement to production.