Webinar Recap: What Changes, What Stays: A Human-Centered View of AI in L&D

AI is rewriting how learning gets designed, delivered, and measured. What it isn’t rewriting is the job of the learning professional.

That was the through line of “What Changes, What Stays: A Human-Centered View of AI in L&D,” a live webinar co-hosted by Yoodli and GP Strategies on September 10. Cara Halter, Senior Director of Global Learning Innovation at GP Strategies, and Mike Rich, Head of Partnerships at Yoodli, spent an hour unpacking where AI adoption in L&D stands, how roleplay fits into learning strategy at institutional scale, and the questions every learning leader should ask before signing a contract.

Halter reviews learning technology for a living. She’s spent 11 years on GP Strategies’ innovation team evaluating dozens of tools in this category, and she opened with a caveat that set the tone: she came in as technology-agnostic as possible. This wasn’t a product pitch. It was a field guide.

Here’s what stood out.

The three phases of AI adoption in L&D

Halter mapped AI adoption into three phases. They aren’t strictly sequential, and moving forward doesn’t mean abandoning what came before.

Phase 1: Productivity and efficiency. You interact with AI. Same deliverables, produced faster: drafting assessment questions, generating scenarios, creating video and audio. Nearly every authoring tool now has an AI-generated button, and most teams have clicked it. Halter’s challenge to teams in this phase: have you redesigned your workflows, or just plugged AI into the old ones?

Phase 2: New learner experiences. Learners interact with AI. This is where AI roleplays, AI coaches, personalized assessment, and just-in-time support live, and it’s where most organizations are actively experimenting today.

Phase 3: Learning in the flow of work. AI becomes invisible. Learning is built into reimagined workflows, with feedback on real performance rather than practice scenarios alone. Rich made it concrete: imagine a coach that knows the call you’re about to take in five minutes, preps you against your organization’s methodology, then debriefs you afterward.

Halter’s shorthand for why this shift is inevitable: ten years ago, learners Googled it. Five years ago, they YouTubed it. Today, they ChatGPT it. Learners now expect real-time, conversational, personalized answers, and learning teams are being measured against that expectation.

Old world, new world: six shifts in learning design

Halter and Rich walked through six paradigm shifts reshaping the field:

1. Discrete courses → continuous learning ecosystems. The course is no longer the unit of value, which raises hard questions about what “completion” even means, and how LMS-era tracking and reporting adapt.
2. Content-centric → context-centric. Information is everywhere. The value is in helping people apply it to their role and moment of need.
3. One size fits all → personalized and adaptive. AI finally makes individualized learning paths feasible at scale.
4. Learning as preparation → learning as ongoing, validated performance. AI can help confirm someone can do the thing, not just that they finished the module.
5. Learners as consumers → learners as contributors and creators. In an AI roleplay or tutoring conversation, the learner drives.
6. Human-designed and controlled → AI-driven, human-governed. Instructional designers shift from writing every word to organizing data, setting guardrails, and pointing AI at the right sources.

Halter, a self-described 30-year learning nerd, acknowledged the discomfort in that last shift. She joked that she’s never met an instructional designer without control issues. Letting learners steer their own experience is genuinely hard, and it’s the job now.

Proof at scale: RingCentral and Ochsner Health

Two case studies grounded the discussion in outcomes.

RingCentral: 90% reduction in certification time. Certifying customer support agents used to consume eight hours of manager time per agent, with managers grading every mock call by hand. Training couldn’t keep pace with hiring, and scripted scenarios went stale. With Yoodli’s AI roleplays built from real support calls, certification dropped from eight hours to one. Managers now open a dashboard, see exactly who’s struggling and where, and spend their time coaching the people who need it. RingCentral scaled onboarding without adding headcount, and new agents reached call readiness sooner.

Ochsner Health: from pilot to system-wide. The Gulf South health system wanted frontline leaders to practice the human conversations that shape patient care: building trust, setting expectations, leading with empathy. Leaders had learned the frameworks but rarely rehearsed them, and there were few tools to coach the coaches across a large, distributed system. Working with subject matter experts, Ochsner built 12 custom AI-scored goals across three conversation types, all mapped to the organization’s values. The biggest gains showed up exactly where baseline scores were lowest, and executives validated meaningful improvement in quality and confidence. The pilot expanded system-wide.

Both stories illustrate the same principle: the organization defines what good looks like, and AI delivers the practice reps and feedback at a scale no human coaching program could match.

AI roleplay is becoming table stakes. A purpose-built platform is the differentiator.

One of the webinar’s sharpest points: the ability to do AI roleplay is showing up everywhere, bolted onto LMSs, content libraries, and video tools. Rich noted that Gartner recently identified AI roleplay as an emerging software category in its own right, precisely because doing it well at enterprise scale requires purpose-built infrastructure, not a checkbox feature.

So the differentiator isn’t whether a vendor offers roleplay. It’s what’s built around it. The speakers offered a due-diligence list for learning leaders evaluating any vendor:

• Is it practice only, or does it include real-call analysis and ongoing skill tracking?
• Who validates the quality of the AI’s feedback?
• Are personas and scenarios customizable to your methodology, or one-size-fits-all?
• What happens to learner data after each session?
• Does pricing hold up at hundreds or thousands of learners, not just a pilot?
• Is the vendor funded and committed to this category for the long haul?

And the trap doors to watch for: roleplay with no coaching depth, no link between practice and on-the-job performance, a glitchy first experience that kills adoption, and vendor lock-in disguised as a platform.

Where humans stay in the loop

Halter and Rich were aligned on this point: the goal isn’t full automation, it’s the right balance of AI scale and human judgment. Halter presented a spectrum of control, from fully human-designed experiences to high-autonomy AI, and made the case that there’s no universally right position on it. High-compliance training belongs further left; exploratory skill-building can sit further right. Her advice: whatever your instinct, take one deliberate step toward more learner-driven design on your next project.

Rich pointed out that Yoodli builds this spectrum directly into the product, with a control that sets how much creative latitude the AI has in any given roleplay. Sales discovery practice might warrant an unpredictable persona; certification for high-stakes procedures demands rigor.

The scale question resolves the same way: deploy AI for reach, speed, and consistency, then layer in human touchpoints like secondary reviews, manager conversations, and facilitated live events where judgment and depth matter most.

The measurement question every CLO is asking

The Q&A’s heavyweight question: how should learning teams measure and communicate program value in this new world?

Halter’s answer was to start with measurement, not end with it. Before designing anything, define the behavior you’re trying to change and what good looks like, then build those rubrics directly into the tool. That turns reporting from smiley sheets and completion rates into behavioral evidence: a team that moved from 35% to 85% on empathy against your organization’s own standard is a story any executive understands.

The second half of her answer pushed further: pair learning metrics with business metrics. Ask your stakeholders how they measure performance, and design the program to connect to that data. Rich extended it into the flow of work: if reps are practicing against a sales framework, run their real recorded calls through the same methodology and let the AI proactively invite them to coaching based on what it finds. That’s the loop between learning the material, practicing it, and proving behavior changed.

What stays

The webinar’s title question got a clear answer. The tools, the delivery models, and the measurement capabilities are all changing fast. The role of the learning professional is not. If anything, learning leaders have a bigger seat at the table, because someone has to define what good looks like, govern the guardrails, and connect practice to performance. Content is still the foundation. Capability is the new deliverable.

---

Want to see what human-centered AI roleplay looks like for your organization? Watch the full recording or talk to the Yoodli team about building practice, coaching, and measurement into one platform.