If the last year of your professional life has felt like one long AI mandate handed down from the top floor — while you quietly Googled “what is a hallucination in AI” at 10 p.m. — first of all, hi. You’re among friends here. Pull up a chair.

We’ve spent a lot of words, as an industry, on two questions: Should we use AI? And how do we roll it out without setting something on fire? The first is settled — AI is here, your people are already using it, and pretending otherwise is like ignoring the smoke alarm. The second already has a good home; if you want a calm, structured, 90-day plan for implementing AI responsibly, we wrote you a whole guide for that.

How AI Is Changing Learning and Development in 2026

The question almost nobody is sitting with is the harder, more strategic one: AI isn’t uniformly good or bad for learning. Its impact splits — sharply — depending on where you point it. And if you’re the person accountable for proving L&D’s value to a CFO who’s watching every line, that distinction isn’t academic. It’s the difference between an investment that compounds and one you’ll be defending in your next budget review.

Here’s the proof that should stop you mid-scroll. In 2025, two rigorous studies looked at AI tutoring and reached opposite conclusions. One, a Harvard randomized controlled trial published in Scientific Reports, found students using a purpose-built AI tutor learned more than twice as much, in less time, than students in one of Harvard’s own excellent active-learning classrooms. The other, a study of nearly 1,000 high schoolers published in PNAS, found that students who leaned on a generic chatbot scored 17% worse on a later, no-AI exam than classmates who never used it at all.

Same technology. Opposite report cards.

The difference wasn’t AI versus no-AI. It was whether the tool preserved the effort that builds the skill — or quietly did the thinking for the learner. Hold onto that, because it’s the spine of everything that follows. As Adam Weber put it in our Performance Paradox conversations, “speed is not a direction.” Point AI at the wrong thing and it’ll just get you to the confusion faster.

A quick gut-check before we go further

Let’s set the scene honestly, because the backdrop is the business case.

Nearly every organization is using AI in some form. Almost none have truly scaled it. And most aren’t seeing real financial return yet — PwC’s 29th Global CEO Survey found that 56% of CEOs reported neither higher revenue nor lower costs from AI in the past year, and only 12% saw both. “AI is everywhere” and “AI rarely pays off yet” are both true at once, a bit like a gym membership in January.

That gap is the whole reason this piece exists. When the average return is this uneven, where you aim AI matters far more than whether you adopt it — and “we deployed AI across L&D” is not a sentence that will impress anyone reviewing your spend. “We deployed it precisely, and here’s the behavior change” is.

For L&D specifically, there’s a quieter problem underneath the hype. We’ve started confusing faster with better. The data is unkind here: for every 10 hours AI saves, teams lose nearly four of them fixing its output, and trust in company-provided AI has actually fallen. Heidi Kirby makes this case with more teeth than we will, in her piece on the double standards AI is surfacing in HR and L&D — worth your time when you’re done here.

So let’s talk about where AI is genuinely working — because the good news is real, and it’s where your next credible win lives.

Where AI Is Improving L&D: Practice, Coaching, and Data

Practice, roleplay, and simulation — go deeper. This is the big one.

Here’s a truth we’ve all known in our bones: people learn by doing, not by watching a video and absorbing it through proximity. Active practice beats passive consumption every time. Carnegie Mellon research found interactive learning produces roughly six times the learning gains of passively watching, and Ebbinghaus’s forgetting curve — undefeated after more than a century — reminds us we shed a brutal chunk of passively-received information within 24 hours.

The strategic problem has always been that practice is hard to scale. You can’t clone your best manager and have her run roleplays with all 400 people on a Tuesday. AI can. And the evidence says it works beautifully — when learners actually do the cognitive lifting. That’s exactly why the Harvard tutor succeeded: it was built to make students work, not to hand them answers.

This is the gap BizReady was built to close. It drops AI-powered roleplays directly inside the BizLibrary lessons your people already watch, so a learner can finish a lesson on coaching or handling an upset customer and immediately practice it — in a realistic, adaptive conversation, with instant feedback, as many times as it takes to feel ready. For anyone measured on leadership-pipeline readiness or time-to-competency, that’s not a feature. It’s scalable practice infrastructure for the exact development you’re already accountable for.

AI in the coach’s corner, not the coach’s chair — yes, with a human in the loop

AI works wonderfully as a sidekick. In an exploratory trial with UK secondary students, AI tutoring supervised by expert human tutors performed at least as well as human tutoring alone — students were a few points more likely to solve new problems afterward. And Stanford’s Tutor CoPilot study found AI assistance helped the least experienced tutors the most, lifting their students’ math proficiency by up to nine percentage points.

Think Robin, not Batman. This maps cleanly onto what David Kelly and Adam Weber keep saying: automate the preparation, never the final decision. AI gets people ready for the moment. Humans own the moment.

The quiet, unglamorous win — “invisible AI”

Not every win comes with confetti — and for anyone wrestling a siloed tech stack, this may be the most valuable section here. As Lori Niles-Hofmann argues in our Performance Paradox report, some of the highest-leverage AI applications are the ones you never see: structuring skills data, surfacing insights, connecting performance signals across systems so people make better decisions. No highlight-reel dunks — just a steadier, smarter operation underneath, and the kind of consolidated intelligence that turns a scattered stack into something you can actually report from. This is what the report calls becoming a “performance architect” rather than a content shop: diagnosing root causes, aligning learning to strategy, using data to inform decisions instead of justify activity. Don’t sleep on it.

Where AI Is Hurting Learning and Development

Now the harder part. We’ll be honest with you, because you deserve it — but we’ll hand you a way forward with every hard truth, because that’s what friends do.

The “just give ’em a chatbot” approach — walk away

Handing learners an unguarded chatbot and calling it development is the trap. Remember that PNAS study? The students using the generic bot looked more capable during practice and ended up less capable when it counted, because the bot became a crutch. It’s the sneakiest failure mode in the whole category — the practice numbers look fantastic right up until the moment they don’t. (Which, if you’re the one reporting those numbers upstairs, is a uniquely uncomfortable surprise.)

There’s a reason for this. When AI does the thinking, it strips out the productive struggle that learning genuinely requires. Kirby calls this the cognitive offloading problem, and she’s right.

Here’s the hand on the shoulder, though: the fix isn’t banning AI. In that very same study, a version of the tutor with learning guardrails erased the harm almost entirely. The lesson isn’t “AI bad.” It’s “design AI to scaffold the struggle, not skip it.”

Content-volume-as-impact — please stop. (We say this with love.)

We need to talk about the content treadmill. Cranking out more courses, faster, padded with AI avatars and synthetic voices, is not the flex we think it is. More learning is not better performance. As Niles-Hofmann puts it, “learning is a tax on employees” — spend it deliberately and sparingly. A well-placed job aid can run circles around a 30-minute course nobody asked for.

And — gently — nobody’s inbox is begging for more uncanny-valley avatars. Kirby said it best: if people already think human-made training is boring, just wait until they meet the AI slop.

The way forward is a redirect. Point AI at your strategy and your broken processes, not your output volume. Use the hours AI frees up on the human work that actually moves the needle — coaching, needs analysis, real conversations with the people you serve. David Kelly’s three questions for L&D in 2026 is the gentlest, most useful mirror we know for this.

Automating the human touchpoints that build trust — don’t

Some moments carry trust: feedback, coaching, evaluative judgment, the final call on who’s ready and who isn’t. Trust is the limiting reagent of this entire enterprise — without it, nothing else works, and shareholder data in the Performance Paradox report suggests it’s measurable in the financials, too. So automate the prep and the pattern-spotting all day long, but keep judgment and accountability firmly in human hands. (Kirby’s double-standards piece walks through exactly what happens when we forget this. It isn’t pretty.)

How to Use AI in L&D Without Undermining Learning

If you take one thing from this, take this: preserve the effort, automate the prep.

AI belongs on the parts of learning that benefit from scale, repetition, and preparation. It should stay off the parts where the struggle is the learning, and where trust gets built between real people.

That reframes the old comfort line we’ve all leaned on. This was never really about whether AI will replace people. It’s about whether we’re intentional enough to point it where effort should be amplified rather than removed. That’s the L&D-sized answer to the Performance Paradox: not faster learning. More intentional learning.

AI in L&D: What to Deepen, and What to Abandon

Stop grading AI pass/fail. Start judging it application by application — deepen what builds genuine capability, drop what merely counterfeits it. That discipline isn’t a side skill anymore; it might be the skill. As Weber reminds us, the organizations that win won’t be the fastest adopters. They’ll be the most intentional designers.

Which is, honestly, good news. Because you don’t need to become a machine-learning engineer to lead through this. You need to stay the calm, credible, deeply human voice in a very noisy room — the one asking “but what problem are we actually solving?” while everyone else is busy being impressed by the demo.

That’s the job People and Development has always done. You’ve absolutely got this.