Part 2 of the six-part series From Course Factory to Capability Function.
Previously in this series: From Course Factory to Capability Function: What Changes When L&D Stops Producing Courses
Most L&D teams are asking how AI can help them produce learning faster. That is the wrong first question.
The right first question is simpler: which of this work should exist at all?
Every L&D function carries work that survives on habit. Reports that get compiled and filed without changing a decision. Intake processes that add weeks of queue time. Compliance modules kept in their current form because that is how they have always been delivered. Vendor coordination that exists because nobody cancelled the contract line. When AI arrives, the temptation is to automate all of it. Faster reports. Faster intake. Faster modules. The function produces more, and the organization gets nothing it did not already have.
That is Step 0, and most teams skip it: examine the work before you scale it.
Start with one service or one team. List the work in plain language: the activity, who it serves, what decision or result it produces, and how much effort it takes. Then put each activity through four dispositions.
First, STOP. Ask what would happen if the activity disappeared. If the honest answer is nothing, stop it. But stop is a governed decision, not a cost-cutting reflex. Check for legal or regulatory requirements, policy obligations, safety and control needs, and downstream dependencies. If any of those is unknown, you do not have permission to stop yet. An unknown obligation blocks the decision.
Second, simplify before you automate. Strip the duplicate steps, the extra approvals, the format variations that exist because five stakeholders each wanted their own version. Automate the reduced process, not the obsolete one. Automating a bloated process just produces waste at machine speed.
Third, decide what technology should do and what should stay human. Repeatable, verifiable work with clear data boundaries is a candidate for automation. Work that depends on judgment, trust, sensitive relationships, or consequential decisions stays human-owned. AI can assist that work. It does not own it.
Fourth, and this is where most redesign efforts die: decide what happens to the capacity you release. Freed time is not value. It becomes value only when it moves to named work, with a named owner, aimed at a named business problem, with a measure and a review date. Until then it is a forecast on a spreadsheet.
That last point needs emphasis, because the economics of L&D redesign are routinely misstated. Separate four things that are often blended into one claim.
Production effort is what it costs L&D to make something. Consumption time is what the organization spends using it. Cash is money that actually leaves or stays in a budget. Realized value is an observed business result.
A 60-minute course consumed by 10,000 employees consumes 10,000 learner hours. That is true whether AI built the course in 20 hours or a team built it in 200. Cutting production time does not reduce consumption time by a minute. It does not save cash unless a staffing level, an invoice, or a budget line actually changes. And it does not create business value unless performance changed and someone can show it.
So when a redesign claims savings, ask which of the four moved. If only production effort moved, you have capacity release. Useful, but only if you redeploy it. If consumption time moved, you gave the organization time back. If cash moved, say so with the invoice to prove it. If none of the four is observed, you have a forecast. Report it as one.
None of this requires a new framework or a maturity model. It requires an inventory, four honest dispositions, and the discipline to report only what you observed. One service, one decision path, one page of output. Then pilot one change, measure what actually happened, and decide whether to continue.
AI will change what L&D does. The teams that get value from it will be the ones that decided what their work should be first.
