AI improves the speed and scale of learning and development content creation, but quality still depends on human judgment, accurate prompting, and deliberate review. Generative AI (technology that produces text, images, video, and audio from natural language prompts) has made it possible for L&D teams and subject-matter experts to create training faster than ever before. What it cannot do is replace the domain knowledge, organizational context, and critical eye that makes training actually useful. This article covers what AI can and cannot do in L&D today, how to evaluate tools worth using, and what the shift means for instructional designers and the teams they work with.
AI in learning and development is augmentation, not automation
AI does not replace the human expert in L&D. It removes the friction that stops them from creating and sharing knowledge in the first place. This distinction matters because it changes how L&D teams should think about adopting AI, evaluating tools, and explaining the technology to skeptical managers.
Patrik Schmitt, Chief Product Officer at Easygenerator, made this clear in a recent webinar on AI in e-learning. AI in learning right now is very little about automation, which is perhaps different from a self-driving car where no human is involved. When it comes to authoring, it is more about augmentation. The human is still the one who knows something and wants to communicate it to colleagues. The question becomes how to make that process more efficient, and how to use AI to do the heavy lifting to make content faster and of better quality.
The reason AI works particularly well in L&D comes down to what large language models actually do. Earlier forms of AI, from recommendation engines to fraud detection systems, were built to analyze patterns in structured data. They worked well for platforms like Netflix or Google, but had little to offer L&D professionals. Large language models work with language, which is the medium through which almost all learning happens. Patrik described this as being “the right technology for the right industry” in a way that blockchain or machine learning never were for the field.
On the practical side, this means AI now handles tasks that used to slow down content creation significantly. Writing a course outline from a document, generating assessment questions, producing voiceovers in multiple languages, creating a video with a realistic presenter: these used to cost thousands of dollars and weeks of production time. AI has compressed both. Patrik noted that a professional training video that once cost $10,000 to produce now costs closer to $10 with AI-generated video tools. That compression does not change what good content looks like. It changes who can produce it, and how fast.
Derek Bruce, Chief Learning and Knowledge Officer at Easygenerator, described the shift from his own experience as a practitioner. AI is not going to revolutionize the L&D space in terms of getting rid of people. It will help people. Outside of a product, the word copilot is probably the best description of how AI can support learners, working alongside people to help give better impact to organizations.
Why AI-generated content still needs a human in the loop
AI output in L&D is only as reliable as the human reviewing it. This is not a limitation that will disappear as models improve. It is a structural feature of how large language models work, and L&D professionals who understand it produce better content than those who treat AI output as finished work.
Talha Faridy, Product Lead for EasyAI at Easygenerator, explained the underlying reason during the same session. The foundation of a large language model is based on probabilities. That means it is very likely for the AI to get things wrong and not have the definitive answer you are looking for. What makes it valuable is the iterative process: you work with it, refine, and build toward the result you want. It should not be taken as a one-shot silver bullet but as something you shape over multiple exchanges.
This played out in a concrete example from the webinar audience. David Scrimshire, a practitioner based in the UK, shared a specific cautionary experience: “Be careful using ChatGPT for statistical analysis for training courses. I always check, and have found several instances of AI generating incorrect statistical analysis of data I provided. ChatGPT apologized and said ‘that’s why we are working as a partnership.’” Sarah Larson reinforced the same point simply: “You must check everything AI gives you for validity.”
Seán, contributing from Dublin, made the same point from a content creation angle: AI is great for content creation, but subject-matter experts are still required to verify the output. That combination, AI for speed and structure, human expert for accuracy and context, is what makes the model work.
Derek framed the broader principle well:
The practical implication is straightforward. AI output should be treated as a first draft, not a final product. The human in the loop is not there to fix AI mistakes in a corrective sense. They are there because the knowledge the training needs to convey lives in their head, not in the model.
How to evaluate whether an AI tool is actually worth using
The best AI features in L&D solve a specific problem. They are not impressive for their own sake. As AI becomes standard across software platforms, the ability to distinguish tools with genuine utility from those chasing a trend will matter more and more.
Patrik gave the clearest test during the webinar:
His framework is worth keeping: identify the real problem first, then evaluate whether a tool materially improves the situation compared to what you were doing before. If the answer is yes and the problem is genuinely costly, the tool is worth testing. If the answer requires squinting, it probably is not.
A practical course design framework shared by David Scrimshire during the session illustrates this well. His approach, developed through working with Easygenerator, breaks the process into four steps: break training into bite-sized sections first, define the learning objective for each section, specify how to test understanding, and only then create the section content. Several attendees responded positively to this as a method because it gives AI clear, bounded tasks rather than open-ended ones, which consistently produces better output.
The same logic applies to leader adoption. Derek described how his team approaches this challenge: “The one thing I’d say is: make it normal, make it accessible, and don’t be scared to break it. I’ve seen leaders who don’t even have LinkedIn profiles. If you think that’s the starting point, then AI is going to be a huge jump. So it’s up to us as a learning team to say: it’s coming, you’ve got to use it, it’s just normal and you can’t break it.”
Structured exposure rather than open-ended exploration tends to work better. Drop-in sessions around specific use cases, writing emails with Copilot, summarizing documents, drafting a learning strategy section, give people enough context to form a real opinion about what AI can and cannot do for them.
One thing that consistently holds people back is prompting. Talha identified it as the most underestimated skill in the space. People go on ChatGPT or some AI tool, start prompting, and when they do not get the response they were looking for, they go back to their normal manual workflow. Investing time in understanding how prompting works changes this. The better you are at prompt engineering, the better your results across all AI tools, not just Easygenerator but ChatGPT, NotebookLM, Synthesia, and others. It is a core professional skill, not a shortcut.
How AI roleplay and adaptive learning are changing skills training
AI makes practice-based learning scalable for the first time. Skills training has always faced the same structural problem: the methods that actually build capability, repeated practice with realistic feedback, are expensive to run at any meaningful scale. One trainer cannot realistically roleplay with a thousand employees. AI changes that constraint.
Patrik described Easygenerator’s approach to this problem. Roleplay is a valuable way to practice soft skills, but the logistical challenge is significant when you have one trainer and a thousand employees. The experiment has been building a product where employees roleplay with an AI that has very specific instructions about how the scenario should unfold and what constraints the learner should run into. Realistic conversations, immediate evaluation afterwards, and a personalized result for each learner.
Two attendees shared concrete implementations that showed how organizations are already building this in practice. Karol Isip, from Telus International in the Philippines, described using AI for agent-trainer roleplay sessions, with AI copilots built to improve learning, productivity, and performance, including documented evaluations and interactive courses with certifications. Stephen Bruington from Gusto in Virginia described using a tool called Call Simulator for AI-based interactive roleplays, and also building the same approach in ChatGPT via custom GPTs or in Gemini using Gems. His observation on the structural value of that approach is worth noting: the value of GPTs and Gems is to fix the structure around the roleplay design and allow sharing and scaling of the solution.
That last point is important. AI roleplay tools are not just a cheaper version of human-led practice. When the scenario is well-designed and the constraints are clearly defined, they make it possible to scale deliberate practice across an entire organization without losing the quality of the feedback loop.
EasyCoach, Easygenerator’s AI roleplay product, is built around exactly this use case. It allows employees to rehearse realistic business conversations, from sales calls to leadership discussions to difficult internal conversations, and receive immediate feedback on their responses. Because it sits inside the same platform where content is authored, L&D teams can connect the practice layer directly to the training they have already built, rather than managing them as separate systems.
On the adaptive learning side, Talha described it as the next major frontier. Learner preferences are very diverse and everyone learns differently. The goal is to create a system that genuinely adapts to each individual: showing content tailor-made for them, scoring them, guiding them, and helping them retain better. Cracking the code on adaptive learning is the next big challenge for personalized L&D.
Adaptive learning is a system that adjusts the content, pace, or format of a learning experience based on how an individual learner is progressing, rather than delivering the same fixed path to everyone. The technology exists in early form today, but designing it in a way that guarantees all learning objectives are covered regardless of the path taken remains an open problem. According to a 2024 analysis by the Brandon Hall Group, organizations that have implemented adaptive learning report a 40% reduction in training time while maintaining equivalent knowledge retention outcomes, though implementations at full scale remain relatively rare.
What this means for instructional designers and L&D professionals
Worth remembering
The instructional designer role is shifting from content builder to learning consultant, and AI is what makes that shift happen now rather than later. This is not a prediction about the future. It is a description of what is already happening.
Talha described what he sees in conversations with practitioners. Many instructional designers are already looking at their role evolving toward consultancy. Subject-matter experts are now being equipped with instructional design knowledge they did not have before, and AI is what closes that barrier. The ID expertise does not disappear; it gets distributed.
Derek confirmed the same pattern from his own team: “The role of what we would call content designers has definitely moved away from just designing content. It’s moved towards coaching individuals to understand AI.”
Louise Puddifoot, an independent L&D consultant, made a similar observation in a separate conversation on L&D trends: “The big shift I’ve seen really is L&D people thinking in terms of enablement rather than just design or development. I’ve even started to see ‘enablement’ featuring quite a lot in job titles now for people in learning and development roles.”
What this means in practice is that instructional design expertise does not become less valuable. It becomes more widely distributed. When SMEs can build a structurally sound course with AI guidance, the instructional designer’s job shifts toward quality assurance, governance, and strategic consultation on what should be built and why. That is a more interesting job, not a diminished one, but it does require a genuine willingness to let go of the production role.
For L&D teams navigating this shift, Derek’s advice is to approach it the same way you would any significant organizational change: make it feel normal rather than exceptional. Build structured exposure for colleagues who are uncertain. Share the mistakes publicly as well as the wins. And accept that prompting is a professional skill worth investing in, not a shortcut.
According to the World Economic Forum’s Future of Jobs Report 2023, AI and machine learning specialists are among the fastest-growing roles globally, while content creation and instructional roles are shifting rather than shrinking. The organizations that navigate this well will be the ones that help their L&D teams develop AI fluency as a core competency rather than an optional upgrade.
