Artificial intelligence (AI) has given the learning market new energy, but not all of it is pointed in the right direction. Providers are racing to show generative AI (GenAI) features, buyers are trying to understand what matters and the media cycle keeps rewarding whatever sounds most futuristic. That creates a familiar market problem. The easiest things to see are not always the most valuable things to buy.
That does not mean AI in learning is overhyped. It means the wrong parts are often overhyped. Content generation, chat interfaces, tutors, assistants and automated recommendations are easy to show in a demo and easy for buyers to understand because they feel practical and immediate. Those capabilities can reduce friction and make learning operations more efficient. But they are not, by themselves, a learning strategy.
The bigger question is whether AI helps organizations build capability, not just produce more learning activity. The next phase of the learning market will not be defined by whether a provider has AI. It will be defined by whether AI improves relevance, quality, trust and measurable outcomes. What is visible in the interface is not always what is valuable in the operating model.
By 2028, digital learning providers will embed GenAI and agentic capabilities as baseline features, while differentiation shifts to content quality, proficiency measurement and governance rather than chat interfaces alone. That is the shift buyers and providers need to absorb. Once AI becomes baseline, the presence of AI stops being interesting. What matters is whether it is connected to the skills, roles, work context, content quality and evidence required to make learning more effective. In many organizations, the learning problem is not a lack of content. It is too much content, too little context and too little evidence that employees can apply what they have learned.
AI-generated learning content can help draft, localize, update and maintain materials. It can help SMEs avoid starting from a blank page and make learning teams more responsive to business change. But without instructional design, source grounding, version control and human review, content generation can multiply the clutter learning and development (L&D) teams have spent years trying to manage. The risk is not that AI will fail to create enough learning. The risk is that it will create more learning than employees can trust, absorb or apply.
The more durable value comes from capabilities that are less flashy but more consequential. Skills alignment, extraction, inference and assessment sit at the center of that value. If AI can identify skills, map them to roles and connect learning to real proficiency, learning becomes more relevant to employees and more useful to the business. AI coaching and reinforcement also deserve more attention. Completion is easy to measure, report and benchmark. It is also a weak indicator of whether someone can perform differently on the job. AI can support practice, reflection, nudges, scenarios and coaching after the formal learning moment ends.
Useful personalization should do more than change the recommended courses on a homepage. It should understand the worker’s role, proficiency, career interests, performance context and skill gaps. It should help employees know what to do next and give managers enough visibility to support development without turning learning into surveillance.
This is where AI in learning starts to look less like a feature race and more like an operating model question. In a prior research note, I argued that AI agents are not valuable because they create another conversational front door. They are valuable when the underlying HR stack is integrated, governed and ready for safe execution. The same applies here. An AI coach, tutor or content generator may look impressive in a demo, but without skills context, validated content, permission-aware data and clear guardrails, it risks becoming another layer of learning noise.
My colleague Mark Smith has made a similar point at the enterprise architecture level. The visible assistant is only the surface. Durable value comes from orchestration, policy enforcement and interoperability that allow agentic systems to operate safely across the enterprise. Learning providers should take that lesson seriously. AI in learning will mature when systems connect skills, content, practice, assessment and workforce context in a governed flow, not when they add another chat window.
That is why governance cannot be treated as a back-office concern. Learning AI needs trusted inputs and explainable outputs. It needs clear content ownership, review cycles, permissions, and controls for hallucinations, bias and intellectual property exposure. In a prior perspective, I explored why trust must be built before dashboards, recommendations or AI-generated insights can be believed. Learning has the same dependency. If the skills library is weak, proficiency signals are inconsistent or content ownership is unclear, AI will not fix the learning ecosystem. It will accelerate whatever quality problem already exists.
This creates a different kind of buying discipline. Buyers should not ask only what AI features are on the roadmap. They should ask which learning outcomes AI improves. Does it identify skill gaps, create better practice, reinforce skills, generate evidence of proficiency, improve content quality and make recommendations that can be trusted? These are harder questions than asking whether a platform has an AI assistant, but they are much closer to the value buyers need.
Providers also need to be honest about visible AI and valuable AI. The market does not need another round of roadmap decoration, and buyers should not reward demo candy that has no clear path to better capability. The market needs proof that AI can improve learning relevance, reduce administrative drag, support capability building and give organizations better evidence of workforce readiness. Simulations, immersive practice, AI tutors and conversational interfaces can all be useful when they are grounded in context and tied to measurable outcomes. The problem comes when the experience becomes the strategy. A clever interface will not overcome a weak skills model. A generated course will not fix poor learning design. A chatbot will not create trust if the underlying content is stale or unvalidated.
AI can help L&D move faster, but speed should not be the end goal. The next phase of AI in learning will not be won by platforms that create the most content or deliver the flashiest interface. It will be won by platforms that help organizations build, prove and sustain capability.
Regards,
Matthew Browm
Fill out the form to continue reading.