Most organizations are using AI in learning in the safest way available to them. They take the training they already run, ask AI to produce it faster, and call the lower price progress.
It is progress of a narrow kind. Cheaper and faster answer an efficiency question, and the standard underneath never changes. If the old training left people partly ready for the real work, the new training leaves them partly ready too, only more efficiently.
The question nobody can answer
Ask a senior leader a direct question about a task that matters in their operation.
Is this person ready to do this task without supervision? How do you know?
The evidence offered is usually one of these:
- A course completion record
- Time served in the role
- A manager’s general impression
- A test taken weeks before the work began
None of that is evidence of competence. It is evidence that something happened near the person. The work itself was never tested. Where an error is cheap, that is a minor weakness. Where it reaches a customer or a patient or a regulated file, it is an unmanaged risk with a training record attached.
What should exist instead
Start from the task, not from a course catalogue.
Write down what good performance looks like for that task. Name the conditions it must hold up under. Name the errors that are not acceptable. Name the point at which a person works alone.
Then, one person at a time:
- Measure the person against that standard using real evidence
- Find the actual cause of any gap
- Give that person the specific help that closes their gap
- Test them again, independently, under conditions that match the real risk
- Keep watching for drift after they pass
The help will not look the same twice. One person needs repeated practice with hard cases they have never seen. Another needs a broken process fixed, so the right action stops being the difficult action. A third needs no training at all: their gap is a workflow step that should never have depended on human memory.
AI makes this practical now. Tailoring a path to one person costs a fraction of what it did. So does repeated practice. So does immediate feedback, and testing before the work is real. Spend those savings on producing the same training for less, and nothing changes in the operation. Spend them on raising the level people are asked to reach, and it does.
Why it does not exist yet
Not because the technology is missing. Standards are argued over, and this forces the argument into the open. Verification is uncomfortable, because it would find experienced people who are not at standard. The performance data sits in systems that learning teams cannot reach. Those are leadership problems. No content tool solves them.
Start with one task
Do not begin with a platform. Begin with one task where errors are expensive and readiness is currently assumed.
Write the standard. Support a small group toward it, one person at a time, each by the route their own gap requires. Verify them independently before they do the work for real.
Measure three things:
- How many were verified ready before they started
- How long it took to get them there
- What their early errors cost, compared with the group before them
If those numbers do not move, drop the idea. If they do, the organization has something it has never had: proof of readiness, task by task and person by person.
Organizations do not need more training produced faster. They need fewer people sent into important work on an assumption.
Dr. Ravinder Tulsiani is an Enterprise Capability Executive with more than 20 years in enterprise learning and workforce transformation, and the author of AI for Learning & Development: Faster Content Isn’t Better Learning.
