A reality check of AI hype in learning design from 100+ practitioners and two experts

If you skim LinkedIn or the 2026 trend reports, you will notice that AI in learning design sounds either like a magic box that can solve all kinds of problems or an absolute mystery that is yet to be discovered. There’s a lot of buzz around agentic AI building your courses, personalization running itself, and how learning designers' roles have evolved with this.

Then you ask actual learning designers.

At the start of our recent webinar with Tim Slade, we had some very interesting discussions, but one question stood out. We had asked the audience, consisting of 100+ learning designers:

In one word, how do you feel about AI in learning design right now?

The chat filled up fast, and the answers were strikingly consistent:

"Overwhelming, but exciting." "Excited, but worried." "Excited, but there are so many tools." "Excited, but there was too much slop." "More cautious because of people running without guidelines."

Notice the pattern!

Nobody is really anti-AI. But also, almost nobody is all-in. People felt more excited but more cautious about AI than they did a year ago.

Tim Slade named the feeling as a duality.

"Professionally, we're excited about the possibilities that AI opens up and enables for us. But personally, there's a lot of hesitation about what AI means for skills, our jobs, intellectual property, the environment and the economy."

That duality deserves better than another hype list. So instead of publishing the usual predictions, we're doing something different: taking the biggest claims circulating in L&D right now and holding them up against what we discussed with practitioners. We have also included Lars-Petter's own field notes coming after 25+ years of building learning, from agency work to co-founding We Are Learning.

Claim by claim. What's true, what's hype, and what it means for your work in learning design.

1: "Agentic AI is replacing the content factory"

The hype: We no longer need to ask if AI can build a course. Agentic AI - unlike a copilot that assists a human - diagnoses training needs, maps objectives, writes content, generates assessments, and deploys finished SCORM packages without a human author in the loop. Some go so far as to declare that traditional instructional design is obsolete for modern enterprises.

The reality check: When we asked practitioners about the AI-generated training they'd actually encountered, they mentioned that it was “full of tells, generic, unimaginative." "It comes off lazy and disingenuous." "Relying on AI helps build faster, but if not done with intention, the result is crap."

During the conversation, LP also mentioned that the output isn’t meaningfully improving.

"It is not magically getting better from what I have seen so far…The skill still lies in how well you understand and can apply learning design. AI can support you, but it doesn't do the work for you."

As a learning designer it lies in your hands to decide what actually needs to change in learner behavior, understand the emotional stakes of a course, and whether a simulation rings true or even ask the harder question: is this the right training intervention at all?

Only when you understand this can you put AI to work to help produce a course.

What this means for you: The content factory isn't being replaced; it's being replicated at speed. If your value is production alone, AI is genuine competition. But if your value is intent, analysis, and judgment, then you have to think beyond AI. It is then simply an output tool, and not the whole package. As one attendee put it: "With AI doing the design, it's opening up more time for analysis."

2: "Hyper-personalization at scale is finally here"

The hype: AI can now create learning paths that adapt to each person's role, skill gaps, prior knowledge, and behavior data, with reports of completion rates jumping 30% when paths are meaningfully personalized.

The reality check: It sort of can…perhaps. But for most tech stacks this is still something for the future. And very often, you don't want it.

There are two points here that a lot of trend reports skip.

First, personalization isn't always the goal. If you're training 4,000 frontline workers to respond to a safety situation in one specific, consistent way, adaptive divergence is a bug, not a feature. Compliance, safety, brand, and procedure training often demand uniformity, and not quite personalization.

Second, today's AI personalization mostly works on the flattest possible content. Adaptive paths tend to assume standardized building blocks — text, quizzes, images — that can be reshuffled per learner. The more immersive an experience becomes, with characters, scenarios, branching dialogue, and emotional stakes, the harder it gets to personalize automatically.

And then there's the bill. There have been instances where an L&D leader at a major tech company projecting a 10x increase in token costs next year, with some in the industry talking about 50x.

Imagine those costs if you have 50,000 employees!

Free-flowing, per-learner AI generation looks very different when the tokenomics land on your budget line.

As Tim framed it: teams used to budget for human capital alone; now it's human capital plus AI token capital, and balancing the two is a new discipline.

What this means for you: The organizations getting real value from personalization did the unglamorous foundational work first — clean skills taxonomies, defined job architectures, meaningful learning objectives. This is exactly what needs to be done, even before you start using AI.

Then, personalize where variation helps learning. Standardize where consistency is the learning. Knowing the difference is a design decision, not a platform feature.

3: "AI simulations are the new default"

The hype: AI-driven coaching simulations and digital humans now make practice-based learning — once a premium option requiring weeks of expert build time — available to everyone, on demand, at scale.

The reality check: There’s a pitfall here, the uncanny valley.

"AI can absolutely create a virtual person you need to interact with," LP notes. "But the issue is that it very often is still a bit uncanny.” We still get AI-generated avatars that don’t quite look real, and not too artificial either, but there’s an eerieness to it that can often distract learners from completely focusing on training.

His broader point from the webinar: as a learner, you shouldn't be able to identify anything as AI-generated. The uncanny valley doesn't just apply to 3D faces — it applies to any AI output that doesn't feel like it was made by someone with intention. When it does feel that way, the whole experience falls flat.

Tim's Waymo test captures the dynamic perfectly.

"When it works really, really well, my tolerance for it goes way, way up," he said, describing the self-driving cars in Phoenix — right before recounting the time one trapped him doing a donut in a gas station. "I was like, get me out of here. I need a human driving this thing."

So where does AI-driven dialogue genuinely work today?

In the use cases built for it.

Language training, where conversational practice with a patient, tireless partner is exactly the point. Presenting to a virtual crowd. Rehearsing structured, high-repetition conversations. In those contexts, AI practice is a genuine breakthrough. But as a blanket default for every scenario produces exactly the flat, creepy experiences that make learners distrust the whole category.

What this means for you: AI can build the structure, write the dialogue, and wire up the branching. But choosing where AI-driven practice belongs — which decisions matter, what's at stake, what good behavior looks like, and where a human or a crafted scenario serves better — is design judgment. Match the tool to the moment, and your learners' tolerance for AI goes way up. Mismatch it, and you've built a donut in a gas station.

4: "Learning designers are becoming strategists and data interpreters"

The hype: As AI absorbs production, designers evolve into "learning strategists, experience architects, and data interpreters who align solutions with organizational goals." Specialist production skills fade; strategy is the future of the role.

The reality check: "A great learning designer — who has always been something of a jack of all trades — now no longer has an excuse for creating boring learning content," his says.

"But even with AI, it still requires creative skills, imagination, playfulness, an eye for detail, a visual mind. For a learning designer these are still key skills that are hard to master. AI does not do this job for you. And if you believe it does, and move over to focus only on strategy and data, then you will not do your job and will not change anything — because the actual content facing your learners will still be boring, unengaging, and flat."

In other words: strategy without craft is just as sloppy.

One attendee shared that their department "was reduced because it was seen as just order takers, and AI is faster."

Tim's warning in the webinar was the same: "If the business just sees us as order takers, eventually they're going to go — I don't need you to be my order taker. AI can be my order taker."

But the answer isn't fleeing production for pure strategy. It's raising the bar on both. Tim again: "The commodity of the future is imagination — who has the best ideas, and who has the best ability to explain those ideas and get to a prototype as quickly as possible."

And LP: "There's no excuse for bad or boring, unengaging learning anymore. It really isn't."

What this means for you: Yes, move upstream — closer to behavior, business outcomes, and evaluation. And keep your craft sharp, because AI just removed every excuse for the click-next course. The designers who thrive won't be the ones who abandoned making things for slide decks about strategy. They'll be the ones whose taste, imagination, and eye for detail direct the AI — and whose learners can feel the difference.

Vibe coding vs. the authoring tool

Vibe coding is genuinely impressive. Both Tim and LP mentioned during the webinar: "If you haven't tried a vibe coding or vibe designing tool, I encourage everybody to give it a spin," Tim said, describing outputs that were "shocking" — things he didn't believe would be possible a year ago.

LP agreed: for the ideas that used to die with "that can't be done with our templates" or "that's a $10,000 custom build," you can now have a working interactive prototype in minutes. As a prototyping and ideation tool, it's already 100% useful.

But the audience's questions cut straight to the unglamorous part:

How do you edit this content a year from now? How do we revisit courses quarterly? What happens when the tool you used disappears? What about accessibility? Where does it even live?

These aren't edge cases. They're the difference between a demo and a learning ecosystem. LP named the practical gaps on the call: vibe-coded output has to be hosted somewhere; large organizations won't accept files "flying around"; many tools still fumble basics like image handling; and security, procurement, and IT reviews ask questions — which models, hosted where, governed how — that a one-off generated artifact can't answer.

Tim's counterpoint is that much of this is promptable.

You can prompt a SCORM wrapper, alt text, captions, tab order — if you know to ask, which itself is a designer skill. And maybe some content becomes disposable: why maintain what you can regenerate? But "just regenerate it" is not a governance strategy when the content encodes compliance requirements, brand voice, localized versions, and a year of stakeholder revisions.

Content built without a plan for ownership, versioning, and updating is content with an expiry date.

What this means for you: The real question is how the vibe coding and course authoring merge: the creative freedom of describing what you want, inside an environment that gives enterprises what they require — inline editing, localization, versioning, hosting, security, and a system that still opens a year later.

That convergence is where we're placing our own bets at We Are Learning, and it's the question worth asking of every tool in your stack: not just "what can it generate?" but "who maintains what it generates?"

The pattern across all of it

Look across these reality checks and one thing is impossible to miss: the work AI cannot do is the work that has always mattered most in learning design.

Asking the right question before building anything. Understanding why behavior hasn't changed yet. Designing a scenario that mirrors real emotional stakes. Choosing the right intervention for the right gap. Evaluating whether anything actually shifted. And — LP would insist we add — making the thing itself genuinely engaging, because your learners never see your strategy deck. They see the content.

These were always the valuable parts of the job. The writing, the formatting, the narrating, the rebuilding — necessary, but never sufficient. AI is absorbing the necessary. What's left is the craft, and the bar for it just went up.

AI won't save bad learning design. But in the hands of thoughtful designers, it might finally help good learning design reach everyone.

Make it matter.