AI for learning design: Good vs. bad use cases in 5 areas

Yes, AI is making it faster to build training.

But it's also making it faster to build bad training.

Every L&D team is asking the same question right now:

"How do we use AI without dropping training quality?"

The answer depends on where you draw the line between production and judgment.

AI is an amazing tool to speed up production and give new enablement tools for learners.

But it makes the designer’s judgment even more important.

Here are 6 areas where AI has good and bad use cases in learning design:

1. Needs Assessment

AI is excellent at processing hundreds of transcripts and surfacing patterns in hours.

But AI can't sit in a stakeholder meeting and hear the subtext, or watch a new hire on the floor and notice which small workflow moment is costing minutes on every call.

That still requires the designer’s judgment.

And if the needs assessment skips it, every design decision that follows is built on the wrong foundation.

Good use: Use AI to analyze customer interactions, call transcripts, and SME interviews to surface real skill gaps.

Bad use: Use AI to generate a needs assessment from a job description.

2. Task Analysis

Task analysis has to weight every task by three factors: risk to the business if done wrong, frequency, and complexity.

AI defaults to surface-level signals.

It ranks frequent tasks higher because they appear more in the data.

But a complex escalation that happens twice a month and carries serious consequences if handled incorrectly needs more training support than a routine task that happens fifty times a day.

So the designer’s judgment is still crucial to focus the training on what actually matters.

Good use: Use AI to organize transcripts, job descriptions, and SME input into a structured task list. The designer validates and weights tasks by impact, frequency, and complexity.

Bad use: Leave task analysis fully to AI and risk less frequent but more important tasks going unnoticed.

3. Scenario Generation

The scenarios AI generates from source material are usually technically accurate.

But real customer interactions aren't that clean.

The customer interrupts, changes the subject, gets emotional, and asks something that falls in a gray area between two procedures.

Those friction moments are where a rep's confidence holds or breaks.

And they only show up in the actual work.

AI can draft the variations.

But if nobody decides which friction points matter most, the reps can handle the textbook version of every call and none of the real ones.

Good use: Use AI to draft ten variations of a customer escalation scenario. The designer decides which friction moments matter first.

Bad use: Use AI to generate scenarios from source material. They cover the content but skip the messy, ambiguous situations learners actually face.

4. Content Development

It’s a common mistake I see when producing training these days.

A 40-page policy document goes in.

A course comes out.

Every section gets a module. Every module gets a quiz.

But nobody makes the hard editorial decisions.

What does a new rep actually need in week one versus week four?

Which concepts need practice, and which ones just need a reference document?

Who are the learners, and what reading level do they need?

Using AI to build training without a designer is like vibe coding.

You get something that works on the surface and breaks the moment it hits a real situation.

The production is fast, but the judgment from an expert is missing.

Good use: Use AI to rewrite source material into clean, learner-friendly language. The designer controls structure, sequencing, and practice.

Bad use: Use AI to turn a policy doc into a full course. No one decides what to cut, what to prioritize, or where learners need practice. The training covers everything and prepares them for nothing.

5. Enablement

AI is a great tool to make learning (and reinforcement) more proactive.

Under time pressure, with a customer on the line, nobody searches a knowledge base.

They guess, they ask the person next to them, or they put the customer on hold.

I've worked with teams that had five or six reactive tools available.

Some AI, some not, some pointing to other tools.

Nobody knew which tool to use at which step.

So first we map each tool to specific moments in the workflow.

Instead of "here are your resources, figure it out," the rep should know: at this step, use this. At that step, use that.

With one client, we even developed an internal AI that follows the live calls to propose actions.

Good use: Build proactive AI tools that reinforce the training and assist teams in real-time.

Bad use: Build reactive AI tools that rely on teams to search for help.

—

The moral of the story?

When you see them all, the pattern is obvious.

Every good use case has AI speeding up specific tasks and giving teams new enablement opportunities.

And every bad use has AI replacing the judgment.

So when AI replaces judgment, you risk entire training programs going to waste.

Yes, AI makes training production faster.

But learning designers make training work.