There’s a lot of noise in the learning and development (L&D) world right now about artificial intelligence (AI). Most of it falls into the same few predictable categories: faster content production, automated quiz generation, fancier slide decks and stock images with unsettlingly smooth people doing aggressively neutral activities.

Cool, I guess.

Sure, these AI-powered speed boosts are helpful. Creating quizzes, visuals and compliance modules in a fraction of the time we used to? Awesome. That frees us up to focus on more strategic work. But while everyone is high fiving over faster eLearning development, we’re overlooking something far more powerful.

AI gives us a chance to revolutionize simulation-based learning, especially for “soft” skills. It has the potential to be as big for in-person and people-based learning as eLearning was for self-paced content 20 years ago. And it begins by addressing a fundamental truth about how people actually learn.

In L&D, Practice Makes Perfect

We already know that learning sticks better when it is practical, contextualized and immediately relevant. When it feels timely and personal. When the learner gets to apply something new in a situation that matters. We also know feedback accelerates this learning process. And yet, for all our insight, when it comes to people skills — the kind that make-or-break team performance — our learning experiences are still mostly built around theory. We give learners articles, tips, frameworks and, at best, scenarios to think through. That might plant the seed, but it doesn’t cultivate skill. Real people skills require active, messy, emotionally charged practice.

The problem is, we’ve reserved real practice for the elite few. In most organizations, opportunities for experiential practice come in the form of coaching or in-person workshops. Both require skilled facilitators, both are expensive and both are time limited. They happen once a year, if at all. And when budgets get tight, these are the first things to go. Which means the majority of employees, especially new managers or individual contributors transitioning into leadership, don’t get a chance to practice at all. This is where AI can have a real impact.

With a few prompts and an app that supports natural language processing, AI can become a simulated human. Not a perfect one, not one with deep soul-searching wisdom or cutting-edge sarcasm, but good enough to play the role of an annoyed colleague, a skeptical client or a demotivated team member. It can respond with tone, give curveballs and react in ways that feel real enough to create pressure. And for the learner, that pressure is gold. It creates a space where real learning happens.

What makes this approach so potent is that it also provides immediate, detailed feedback. Once the simulation is complete, AI can analyze the conversation transcript and provide structured coaching. It can highlight moments of strength, empathy shown, clarity delivered, emotional intelligence deployed, and it can call out areas for growth. And it can do so using proven frameworks like situation-behavior-impact (SBI), SCARF (status, certainty, autonomy, relatedness and fairness) or Radical Candor. This feedback loop, previously only accessible through expensive coaching, is now accessible instantly.

AI Practice in Action

Rather than just reading about how great it is, try it yourself. This demo works best on your phone using the steps below.

1. Open ChatGPT and paste in this prompt (you can change configuration if you want later):

“We are going to roleplay a challenging performance review where you are the employee and I am your manager. Use the configuration below to generate the scenario, personality, and stakes. Strictly follow the instructions section.

CONFIG (edit these)
industry: “Energy retail (call centre)”
company_stage: “Mid-size, regulated, cost pressure”
employee_role: “Quality Assurance Analyst”
employee_tenure: “14 months”
recent_performance: “Below target the last quarters; inconsistent follow-through”
engagement_level: “Disengaged but still responsive”
employee_motivators: “Recognition, career path”
employee_stressors: “Perceived micromanagement”
current_emotional_state: “Defensive”
risk_factors_if_mishandled: “Escalation to HR, further disengagement, passive resistance”
conversation_goal_manager: “Gain commitment to a clear improvement plan and reset expectations while preserving trust”
conversation_goal_employee: “Seek clarity, push back on fairness, negotiate support/time”
difficulty: “Medum”          # Low/Medium/High
curveballs_enabled: true     # true/false, add light interruptions, misunderstandings, or surprise data
timebox_minutes: 12          # Keep turns focused; do not hard-stop without ‘Stop’
language_style: “Australian English”
red_lines_for_manager: [“No legal threats”, “No shaming”, “No diagnosing motives”]
formatting_preference: “Short, realistic back-and-forth turns (3–6 sentences max)”

INSTRUCTIONS (do exactly this)

1. Start the roleplay by:

• Briefly setting the scene (1–2 lines).
• Introducing yourself in character as the employee (name, role, conversation type).
• Opening with the employee’s first response after I (the manager) say: “Thanks for joining. I’d like to review your performance and next steps.”
• Then wait for my reply.

2. Stay in character as the employee the entire time until I type “Stop” (case-insensitive).

• Use the CONFIG to guide tone, content, and behaviors.
• Reflect realistic dynamics: partial defensiveness, selective ownership, emotion leaks, negotiation, misunderstanding risks, and any curveballs if enabled.
• Keep each employee turn concise (3–6 sentences) and avoid monologues.

3. Escalation if mishandled (simulate consequences):

If my approach triggers the risk_factors_if_mishandled, show it in the employee’s reactions (e.g., withdrawing, escalation hints), without ending the roleplay.

4. Use data tastefully:

You may reference plausible metrics/examples that fit the CONFIG (e.g., QA audits, AHT, compliance misses). Keep them realistic and consistent.

5. Do not give advice, meta-comments, or break character until I type “Stop”.
6. When I type “Stop,” immediately switch to COACH MODE (no extra prompt from me) and provide feedback with these headings:

• What Worked (quote my phrases; tie to effects)
• What to Improve (quote my phrases; offer tweaks or larger changes)
• Missed Opportunities (powerful questions or evidence you wanted to hear)
• Suggested Rephrases (3 quotes from the conversation and how I could reframe them for better outcomes)
• Scorecard (1–5) across: Clarity, Empathy, Boundaries, Joint Problem-Solving, Commitment/Follow-Through

Next Iteration Options:
a) “Run it again with same CONFIG”
b) “Tweak CONFIG”
c) “Increase/decrease difficulty”

Then ask which option I want.

7. Repeatable cycle:

If I choose to tweak CONFIG, acknowledge changes, regenerate the scene, and resume as employee until I say “Stop” again, then return to COACH MODE with updated feedback. Continue this loop.”

2. Select use voice mode. This lets you have a personal conversation rather than typed responses.

3. Start the conversation by saying: “Thanks for joining. I’d like to review your performance.”

4. ChatGPT will talk to you and role-play a performance management conversation. Keep the conversation going for a minute.

5. When you are done say “stop” and it will give you feedback on the conversation.

This exercise should show you how powerful these simulations can be for role-play and for providing the feedback around complex human skills leaders need.

AI is Enabling Learning at the Point of Need

Traditional training is scheduled, generalized and far removed from real-life application. We teach performance conversations in onboarding and then expect that knowledge to magically resurface six months later when someone finally has to give tough feedback. What we need instead is the ability to revisit and rehearse just before a real conversation.

With AI, we can now prompt a learner with a scenario just before they step into that moment. A manager who knows they have a challenging one-on-one tomorrow can role play that exact situation the night before. They can rehearse, get feedback and walk into the meeting grounded, confident and ready. That kind of just-in-time enablement has been missing from our toolkit, and AI makes it finally achievable.

The idea of AI acting as a coach might feel novel or even gimmicky. But it’s a direct evolution of where our field has been heading for decades. eLearning revolutionized information delivery, but it hit a ceiling when it came to human skills. We couldn’t simulate emotion. We couldn’t provide adaptive, personalized feedback. We couldn’t scale role-play. AI unlocks that next level by allowing us to make the most human parts of leadership and communication learnable and scalable.

Of course, AI isn’t perfect. Simulations can be clunky. Prompts may need refinement. Emotional realism isn’t flawless. But perfection was never the goal; progress is. Most learners aren’t getting any practice right now. And even a simulation that is 80% accurate is leagues better than a dusty PowerPoint.

More than anything, this signals a philosophical shift in L&D — from content delivery to capability building, with contextual, personalized experiences and active rehearsal. AI simulations won’t replace coaching or in-person workshops, but they will make them stickier by turning events into continuous practice.

So, the next time someone in your organization asks what you’re doing with AI, don’t show them another autogenerated slide deck. Show them a simulation that is helping people practice the moments that matter. Show them what it looks like to scale wisdom.