Outsourcing the Thinking: The Real Risk of Letting AI Design Your Training

Ask an instructional designer today to build a course, and there’s a good chance the first draft doesn’t start on a whiteboard anymore. It starts with a prompt. Feed a subject matter expert’s notes into an AI tool, ask for a script, a set of quiz questions, a few scenarios — and in minutes you have something that looks like a finished course.

That’s not the problem. AI-assisted drafting is a genuine gift to this industry, and I use it constantly. The problem is what happens next: for a lot of teams, nothing. The AI draft gets narrated, published, and shipped, and nobody in the process ever had to think hard about the learner.

After 22 years building training for federal agencies and Fortune 500 companies — and after a year and a half of using AI tools daily in my own production pipeline — I’d call this the next generation of click-next training. Different mechanism, same failure: content that was assembled instead of designed.

What full delegation looks like

You can usually spot it fast. The scenarios all resolve the same way a stock scenario would. The wrong-answer feedback is generic — “That’s not quite right, try again” — instead of showing what actually goes wrong when a real manager makes that real mistake. The examples feel like they could belong to any company, because they were built from a prompt that could describe any company.

Nothing about it is broken, exactly. It’s just nobody’s. No one made a hard call about which mistake the learner is most likely to make, or which detail would make a scenario land for people in this specific job. The AI didn’t skip that step because it’s incapable of it — it skipped it because nobody asked it to, and asking well is itself a design skill.

How it happens

Nobody sets out to build a course with no thinking left in it. It happens for reasons that feel completely reasonable in the moment:

• The deadline eats the review cycle. A first AI draft is fast enough that it’s tempting to treat it as the last draft too. The revision pass where a designer would normally ask “would this actually happen to our learners?” gets skipped because the schedule doesn’t have room for it anymore.
• Fluency gets mistaken for quality. AI-generated text reads smoothly and confidently, every time. That confidence is persuasive — it’s easy to mistake “this sounds finished” for “this is correct, specific, and true to how the job actually works.”
• Nobody owns the judgment calls. A prompt can generate five scenario options. Someone still has to know the audience well enough to pick the right one, or reject all five and describe a sixth. When that step gets skipped, the AI isn’t the one who failed — the person who stopped exercising judgment did.

Why it matters more than a bad course

A generic course is a bad outcome on its own. But the deeper risk is what happens to the people building it. Creative and instructional judgment is a muscle. Every time a hard design decision gets waved through to whatever the AI produced first, that muscle gets a little weaker — for the designer, and for the organization’s ability to tell good training from adequate-looking training.

That’s the trap. AI is exceptionally good at producing plausible work fast, which is exactly what makes it dangerous to lean on for the parts of the job that were never about speed. Knowing which scenario will actually change behavior isn’t a speed problem. It’s a judgment problem, and judgment doesn’t improve by being delegated away.

What the alternative looks like

The fix isn’t avoiding AI. It’s being deliberate about which half of the job you hand it:

• Let AI handle assembly, not decisions. Narration, first-draft scripting, formatting, alt text, translation — AI is faster and often better than a human at these, and there’s no creative cost to delegating them.
• Keep the “who is this learner” work human. The specific mistake a new manager makes in week three, the exact phrasing that makes a compliance scenario feel real instead of hypothetical — that comes from having sat with the audience, not from a prompt.
• Treat every AI draft as a first draft, not a final one. Read it like you’d read a junior designer’s work: where is this generic, where does it dodge the hard version of the scenario, where would our actual learners see through it?
• Use AI to explore more options, not to skip picking one. Ask it for five versions of a branching scenario instead of one — then do the human work of choosing, combining, and sharpening. More raw material should mean better judgment gets applied, not less.

The one question that reveals everything

If you’re evaluating a course — yours or a vendor’s — ask this: “Which decision in this course could only have been made by someone who understands our learners, and which parts could have been produced for any company in America?”

If the honest answer is “not much,” the thinking got outsourced along with the drafting. If there’s a real answer — a scenario detail, a piece of feedback, a choice about what to leave out — that’s evidence someone was still doing the job AI can’t do.

At DGS Designs, we use AI throughout production because it’s a genuinely great tool — and we’ve spent 22 years making sure the judgment behind every course still belongs to a person who knows the learner. Contact us if you want training built by both.