Flat AI access is the most expensive way to adopt AI

Flat access hides important differences

People arrive with different work, experience and levels of confidence. One person may be exploring AI for the first time. Another may have developed a repeatable workflow that saves hours of skilled effort. A third may be ready to let an agent take a bounded action on their behalf.

Treating those situations as identical is simple to administer, but it does not help any of them particularly well.

It can also produce predictable waste. When the system gives no useful signal about cost or fit, people naturally choose the most capable model available. When training is optional and disconnected from access, knowing how to work efficiently brings no practical benefit. When a fixed allowance runs out during useful work, the official route becomes a wall rather than a service.

That last point matters. People do not usually look for an unsanctioned tool because they want to avoid governance. They do it because they still have work to finish. If the supported path stops being useful, policy alone will not keep people on it.

Let capability be earned

Earned capability starts everyone with a supported baseline and lets access grow in graduated steps.

As someone demonstrates that they can use AI effectively, efficiently and responsibly, they can earn access to more capable models, a larger budget or more autonomy. Progression is based on demonstrated capability, not seniority or job title.

This is not a one-time course followed by permanent access. Capability changes as tools, risks and working practices change. The evidence should therefore be blended and reviewed over time:

• useful training that teaches practical ways of working;
• demonstrated application to real work;
• peer, mentor or manager judgement;
• operational signals that help explain efficiency and risk;
• a clear route for exceptions when the work genuinely needs one.

No single score should decide whether someone is capable. A usage number can be incomplete, easy to game and uncomfortably close to surveillance. Measurement should inform a human judgement and help people improve. It should never become a leaderboard for shaming them.

More than model access

Capability has several dimensions.

Model access is the most obvious. Some work needs a frontier model; much of it does not. People who understand the difference can make better choices without a blanket mandate forcing everyone towards either the cheapest or the most powerful option.

Budget should reflect useful work and demonstrated efficiency. A capable person should have a supported way to continue valuable work instead of meeting an arbitrary wall.

Autonomy becomes more important as AI moves from conversation into action. Reading information, drafting a response and changing a production system carry very different consequences. The permission to let an agent act should grow with the person's capability, the quality of the controls and the risk of the action.

These dimensions do not need to move together. Someone may be trusted with a larger budget for analytical work while still requiring human approval before an agent takes an external action. The point is to match capability and control to the work.

Platform before people

Earned capability is only fair if the organisation fixes the waste it controls first.

A person should not lose access because the platform routes work inefficiently, hides cost, applies unnecessary overhead or fails to provide a suitable supported option. Those are platform problems. They should be solved centrally before individual behaviour is assessed.

The same principle applies to the governed path. It must remain capable enough to do real work. People need useful baseline access, timely exceptions and a supported route to continue when a legitimate need exceeds their normal allowance. Governance succeeds when the safe route is also the practical route.

Why this can accelerate adoption

The word earned can sound like a brake. Used well, it is an accelerator.

It gives people a reason to develop practical skill because greater capability follows. It directs expensive tools and budget towards work that can use them well. It creates a supported alternative to the hard limit that drives people elsewhere. It also makes the organisation's investment in training, platform engineering and governance part of the same adoption model.

The organisation gains more predictable spending and a clearer relationship between access and value. The individual gains a route to better tools, more room and greater autonomy. Those outcomes reinforce one another.

This is governance as enablement. The controls exist to help capable people go further, safely, and to help everyone else see how they can get there.