From content to conversation: Learning that adapts in real time
Explore how agentic systems can create learning experiences that are as responsive as a teacher, as patient as a mentor, and as available as the moment curiosity strikes. Monica Arés builds learning systems at the intersection of emerging technology and human capability, from platform-based learning at Amazon to the first immersive learning ecosystem at Meta, and now agentic learning at Microsoft. She focuses on designing systems that don’t just deliver information but actively develop capability, curiosity, and real-world skills. For decades, digital learning has been built for scale. Knowledge was packaged, delivered, and made accessible to millions. But the things that make learning actually stick—curiosity, trial and error, and learning with others—were never easy to build into a platform. Not because they didn’t matter, but because the technology to support them at scale didn’t exist. The result is a system that treats learners like visitors in a library rather than participants in a process. It’s useful if you know exactly what you’re looking for, but it’s silent when you’re stuck. Even the most promising AI tools have struggled with this. When Sal Khan introduced Khanmigo, the idea was simple: give every student access to an AI tutor. In practice, usage was more limited than expected. As he described in Why Sal Khan’s AI revolution hasn’t happened yet, according to Sal Khan, the AI sat in the back of the room waiting to be asked for help. Some students engaged. Most didn’t. Research, such as that reported in The Course Is Dying as the Unit of Learning, shows that only 10–20% of formal training translates into lasting behavior change. Not because people don’t want to learn, but because the systems weren’t designed around how learning actually works. This recognition is already driving change across the industry. The most promising initiatives are rethinking learning from the ground up, built around demonstrated competency rather than content completion. Learning doesn’t happen to us. It happens between us. It’s a conversation. The opportunity now is to move from tools that wait for input to systems that participate. Agentic systems shift the role of technology from passive responder to active partner. They observe, adapt, and respond in context. They know when to step in, when to challenge, and when to guide. For anyone designing learning: Stop asking what content to deliver. Start asking where conversation should happen. What learners actually want from AI This shift isn’t just about better technology; it’s about strengthening the human skills that make learning stick. I saw this clearly in an early experiment at Imperial College, London. My team and I took a full university course—lectures, readings, and assignments—and built a system that could speak to all of it. It was multimodal, so when something was relevant, like a slide, a video clip, or visualization, it surfaced directly in the interaction. Students didn’t use it during class or for assessments. Instead, they used it as endless office hours—at 2 AM when curiosity struck, when they were stuck on something specific, or when they had a question they weren’t comfortable asking during the lecture. Many used it to go deeper on their own projects, following threads that the course itself didn’t have time to explore. Students told us it was the most effective way they had learned. Not because it replaced the professor, but because it was there for all the moments the professor couldn’t be. That pattern holds at every level. Learners don’t want more content. They want something that meets them where they are, connects to what they’re trying to do, and adapts as they grow. When I joined Microsoft, my team began asking the same questions, this time at a very different scale. Why content alone isn’t enough Most learning platforms are curated and linear. You put content in, and you know exactly what comes out. In agentic systems, learning is no longer a line. It’s a loop. The system responds to learner inquiry, adapts its behavior, and changes what happens next. Every conversation is shaped by who the learner is, what they want to build, and where they’re stuck. This isn’t a smarter playlist. It’s co-intelligence. The learner is no longer just consuming what the system delivers. They’re shaping what it becomes. The question worth asking is whether your current tools deliver content or develop thinking. Evolving how we approach video experiences Video is one of the most widely used formats in learning. It’s engaging, but fundamentally one-directional. When you make video conversational, everything changes. My team at Microsoft started working with tools like Synthesia, an AI video platform, to add layers of AI to different components of the video experience. The first unlock was avatars as a new medium for delivery: faster edits without revisiting the greenroom, instant localization, and script updates without re-recording. But that was just the beginning. The next layer used AI to break the video apart by topics and demos and to create new modalities, like podcasts and recaps. This gave users options to explore the topics they cared about, in the order that made sense to them. Then we went further through an internal pilot: using knowledge bases and intelligent agents turned video into a two-way conversation. Learners can now stop a video at any point and ask a question, brainstorm, go deeper on a topic, role-play, or explain what they just learned in their own words and get feedback. What comes next The shift goes deeper than video. This evolution is currently happening along two paths. The easiest place to start is adding layers of AI into content we already have, bringing choice, conversation, and adaptability to existing formats. The other path builds fully agentic systems from the ground up, designed to observe, process, and evolve alongside the learner. Both are important. One lowers the barrier to entry for organizations. The other points to where this is ultimately headed. Looking ahead, two emerging capabilities will take this even further. The first is AI-powered development environments, where learners can build what they imagine. Not just describing a solution, but prototyping it. Creating working systems, simulations, and tools that demonstrate thinking in action. When you can build what you’re learning, the artifact itself becomes the evidence of understanding. The second is screen-aware AI. A system that can see how you build and solve problems in real time becomes something more than a tutor. It becomes a co-intelligence, adapting to your context, assessing your thinking, and measuring proficiency through observation rather than recall. Context becomes the curriculum. Building for agency, not dependency For a long time, scale meant standardization. Reach meant letting go of depth. Those constraints are lifting. For the first time, it’s possible to build learning systems that are as responsive as a teacher, as patient as a mentor, and as available as the moment curiosity strikes. Systems that don’t just deliver knowledge but also adapt and grow with the people using them. The goal isn’t to create technology that people depend on. It’s to create technology that makes people more capable. More curious. More resilient. More able to think and build on their own. That isn't just better technology. That is dignity. Because learning was never about transfer. It was always about transformation. Where to start If you’re building or leading learning: Take one piece of content you already have. Identify where learners typically get stuck. Turn that moment into a conversation. If you’re looking for a place to begin, go to AI Skills Navigator, an agentic learning space that brings together AI-powered skilling experiences and credentials to help individuals build career skills. Be sure to check out the skilling sessions there. Watch Monica Arés’s Synthesia presentation on demand, as she explores agentic learning systems and what comes next.
