Whatever your L&D team has decided about AI, your workforce has already voted. Microsoft and LinkedIn's Work Trend Index found 75% of knowledge workers were already using generative AI at work in 2024, and 78% of those users were bringing their own tools without waiting for the organisation's permission, training or guardrails. The L&D profession is moving too: LinkedIn's Workplace Learning Report found 71% of L&D professionals exploring or already integrating AI into their work.
So the strategic question was never "should we let AI into our learning?" It's already in. The real question is the practical one: which jobs should we hand it, and which jobs will it quietly botch if we do? Having spent the past few years using these tools daily while building learning content, here's our honest scorecard.
What AI is genuinely brilliant at
The blank page problem. First drafts, outline options, alternative framings of a tricky concept: AI produces in seconds the raw material that used to consume the first day of a project. The productivity research is real (a controlled study in Science found professionals with AI assistance worked 40% faster with measurably better output), and in learning design, where every module starts as a pile of SME content and a deadline, that speed buys the thing projects never have: time to think.
Variation at scale. Ten scenario variants for different roles, the same concept explained three ways for three audiences, a bank of retrieval questions from a finished module. This is mechanical-creative work that used to be skipped for budget reasons, and its absence is exactly why so much training feels generic. AI makes the tailored version affordable.
Patient, judgement-free practice. Role-playing a difficult conversation, drilling product knowledge, asking the "stupid" question a learner won't ask a facilitator: a well-prompted AI is endlessly available and never sighs. Practice and retrieval are the most evidence-backed ingredients in learning, and AI lowers their cost to nearly zero.
A second pair of eyes. Readability checks, jargon spotting, "what's confusing about this explanation?" critiques. Used as an editor rather than an author, AI strengthens exactly the quality pass most projects never budget for. (Our own in-house tools sit in this camp: review and support, with a human holding the pen.)
Where it falls over
Facts, confidently wrong. AI fabricates. In a marketing draft that's embarrassing. In compliance, health and safety or clinical training it's dangerous, because the errors arrive fluent and assured. Every fact, figure, legal threshold and procedural step in AI-drafted learning content needs human verification against the source document. No exceptions. The fluency is precisely what makes the errors hard to spot.
Your organisation's actuality. AI knows how restructures work in general. It doesn't know your org chart, your culture's unwritten rules, what happened last time, or why the warehouse team will read that sentence differently. Learning content earns trust through specificity, and generic-but-polished is the house style of unedited AI. Learners detect it faster than you'd hope.
The judgement about what to leave out. Learning design's hardest decisions are subtractive: what the learner doesn't need, which the research says matters enormously (cutting extraneous material improved learning in 13 of 14 controlled tests). AI is an including machine. Ask for a module on data privacy and it will dutifully cover everything adjacent. The editorial spine, this matters, this doesn't, this goes first, remains stubbornly human work.
Your team's own skills, if you're careless. Microsoft and Carnegie Mellon's study of knowledge workers found that higher confidence in AI was associated with less critical thinking applied to the task. For L&D teams the irony writes itself: the people responsible for the organisation's capability quietly letting their own atrophy. The fix is role clarity: AI drafts, humans decide, and the deciding is the job.
The sensible posture
Treat AI as the most capable junior team member you've ever had: fast, tireless, widely read, occasionally wrong with total confidence, and unaware of office politics. You'd never let that person ship to the whole organisation unreviewed. You'd also never refuse the help.
The teams getting this right aren't the most enthusiastic or the most resistant. They're the ones with the clearest editorial standards, because AI raises the volume of plausible content, which raises the value of knowing what good looks like. That judgement, applied to learning content, is what we've built our practice on, with AI in the toolbox and a human hand on the pen.
The tools will keep improving. The judgement keeps its job.