Part 6 of the six-part series From Course Factory to Capability Function.
Previously in this series: Release Is Not Value: What to Measure When You Redesign L&D
After five installments on what changes, here is what the changed function looks like from the inside. Not a vision statement. A description of the work, the roles, and the operating rhythm.
It is smaller in production and larger in judgment. The course-building team shrinks, because AI handles first drafts, routine updates, and standard production work. What grows is the diagnostic and advisory capacity: people who can walk into a business unit, find the real performance gap, and say what will fix it, even when the answer is not training. The job titles may stay familiar. The day-to-day work does not.
Intake changes first, because intake is where the factory model is most visible. Requests no longer arrive as training orders. They arrive as performance problems, and the function’s first move is diagnosis: the four questions from installment 3, answered in writing, before anything is built. What is the observable gap? What evidence points to the cause? What else could explain it? What would show the fix worked? Some requests survive this and become training. Many become recommendations to fix a process, an incentive, a tool, or an expectation. The function gets comfortable saying no to training requests, because saying yes to the wrong training is what made it a factory in the first place. The diagnostic write-up takes a day. Building the wrong course takes months.
The work mix follows the allocation from installment 4, and it is reviewed, not assumed. Machines run the repeatable operations: scheduling, tagging, routing, compiling completion data. AI copilots accelerate design and analysis, with human verification built into every workflow as a named step, not a hope. Agents handle bounded sequences like adaptive onboarding paths, with explicit guardrails on what they may do, what they must escalate, and where the audit trail lives. Humans own diagnosis, stakeholder judgment, and every consequential decision. Escalated humans own the exceptions the other four push up. Twice a year, the team revisits the allocation and writes down what changed. The document is short. The discipline of maintaining it is the point.
Measurement is built in, not bolted on. Every significant investment carries the one-page plan from installment 5: the decision it serves, the evidence that would show it worked, the source of that evidence, and the review date. The function’s standing report to the business shows performance evidence with the four-way separation intact, not activity counts. When leaders ask what L&D did this quarter, the answer is what changed in the business, stated plainly, with limits acknowledged. The first time a function reports “we stopped this program because the evidence did not support continuing it,” something shifts in how the business sees L&D. That is the moment the redesign pays for itself.
Governance is lighter but stricter. Lighter, because Step 0 from installment 2 cleared the work that existed from habit rather than need. The portfolio is smaller, so oversight can be sharper. Stricter, because every remaining commitment names an owner, a measure, and a review date. A quarterly review asks three questions: what did we stop, what did we prove, and what did we learn. Anything without an owner or a measure does not survive the quarter. This is not bureaucracy. It is the mechanism that keeps the function from drifting back into factory habits, which is the natural direction of drift.
The team itself looks different. Fewer instructional designers producing courses on schedule. More performance consultants who can diagnose, more analysts who can work with business data, more advisors who can tell a senior leader their training request will not fix the problem and be believed. Hiring changes. Development changes. The people who thrived in the factory may not thrive here, and that conversation needs to happen honestly and early.
None of this happens in one reorganization. The practical path is the one this series has followed throughout: pick one service the business cares about, run it through the full sequence from Step 0 through measurement, report what actually happened including what did not work, and then expand. One service, one decision path, one page of output. Then the next. Each cycle builds the evidence and the credibility for the one after it. Three cycles in, the function has a track record. Five cycles in, it has a reputation.
The series ends where it started. The umbrella over all of it is simple: solve the right problem. A capability function is what L&D looks like when it does that consistently, refuses to do otherwise, and can prove the difference.
