Nearly every organization has run AI awareness sessions and nearly every technology team is operating at an architectural level. The space between them is empty, and it is the only space where an employee actually changes how a real task gets done.
The four levels
| Level | What it is | Who has it | What it produces |
|---|---|---|---|
| 101 · Awareness | What these systems are, how they fail, what a prompt is, what not to put in one | Nearly everyone. Often several times. | Vocabulary. Very little behavior change. |
| 201 · Application | Using AI on the real tasks of a specific role, with that role's actual artifacts and constraints | Almost no one, taught formally | Hours returned. The first measurable value. |
| 301 · Orchestration | Managing AI as capacity: delegating, verifying, and redesigning the workflow around it. Knowing what must never be delegated. | Almost no one | Changed capacity. What a team can now take on. |
| 401 · Architecture | Model selection, infrastructure, integration, data residency, security, governance, audit | Technology teams | Safety and scale. Usually pitched over the room. |
Why the middle is missing
- 101 is easy to deliver at scale. One session, any audience, no preparation specific to a role. It is the level that fits a staff meeting.
- 401 is easy to justify. It is a technical requirement with a named owner and a budget line.
- 201 and 301 are expensive because they are specific. They require knowing the actual artifacts of a role. A program specialist's guidance document. An HR analyst's structured interview. A teacher's unit plan. You cannot build 201 generically, which is exactly why generic providers do not build it.
RAND found district-provided AI training rose from 23 percent to 48 percent between fall 2023 and fall 2024, with a 67 percent to 39 percent split between low-poverty and high-poverty districts, and nearly all of it optional and one-off. That is 101 delivered twice, not a ladder.
What 201 looks like when it is done properly
- Built around the participant's real artifacts, brought to the session, not a sample.
- Time-bounded to a task they do weekly, so the value is visible in the same week.
- Explicit about verification. Every output is checked against a source the participant already trusts.
- Honest about failure modes for that specific task, not failure modes in general.
- Ends with a changed workflow, written down, not a certificate.
What 301 looks like
- The participant maps their own role and decides what gets delegated, what gets assisted, and what stays entirely human, with a reason for each.
- They design the review step, because the review step is where professional judgment now lives.
- They redesign one process end to end, not one task.
- They can explain the boundary to someone else, which is the real test. A boundary you cannot explain is not a policy, it is a preference.
The sequencing rule
Do not run 201 before the adults have time. The single most common failure is scheduling capability building into hours that do not exist and then concluding that staff are resistant. They are not resistant. They are fully allocated. Time is the resource this work actually consumes, and leadership is the only party that can release it.
This is the practical meaning of the claim that technology sits at the top of the pyramid rather than the base. Relationships and adult culture hold it up. Pedagogical practice sits on that. The tool is last, and a rollout that inverts the order fails in a way that gets blamed on the tool.
← Back to the brief