AI-powered learning – Experience Vibe-Learning

Can you define "Vibe Learning" in relation to "Vibe Coding"?

Jan Foelsing: To better understand Vibe Learning, it helps to look at the origins of Vibe Coding. In February 2025, AI researcher Andrej Karpathy came up with this term. The idea: Using today's AI models, even people without programming skills can build functional software by simply describing what they want. AI generates the code, and you iterate through a dialogue. This puts you in a state of flow, and the code itself becomes secondary. You don't need to know how to program; you just need to know what you want to achieve. The Collins English Dictionary even chose "Vibe Coding" as the Word of the Year 2025.

Vibe Learning applies this logic to learning. The starting point is not a curriculum or a predefined course, but a question, an impulse, genuine curiosity. In dialogue with AI, you approach a topic in an exploratory and self-directed way—playfully, with immediately visible results. In an ideal scenario, this also leads to a state of flow. Instead of acting as a lecturer who simply delivers knowledge, AI becomes an adaptive learning partner: it asks follow-up questions, corrects misconceptions, offers alternative perspectives, and adapts to the learner’s prior knowledge and thinking style.

Here’s an example: Instead of booking a two-hour course on agile methods, a project manager can ask the AI, "Help me understand Scrum. First, ask me what I already know about it, and then help me deepen my understanding." The AI doesn’t start with a definition, but with follow-up questions. It builds on existing knowledge, provides relevant analogies, and helps plan the first application. Within 20 minutes, the project manager has gotten an understanding which is sufficient for the next step.

Important is what Vibe Learning is not: It is not a didactic model, not a new learning theory, and not a substitute for in-depth engagement. This approach certainly cannot replace Scrum Master training, but perhaps it helps, for example, in preparation—to bring participants to a common level and get them engaged from the very start? It can best be described as a conceptual framework. An attitude—or, to use the modern term, a mindset—that describes how AI can radically simplify the introduction to new topics. What comes next—reflection, transfer, and deepening—remains a human task.

Who benefits from a learning process that takes place in a playful, interactive dialogue with an AI?

Jan Foelsing: In principle, Vibe Learning is suitable for anyone who wants to explore a new topic or delve deeper into a specific area. The barrier to entry is virtually non-existent: no lengthy course, no set schedule, and no preparation are required. You start with a question and are quickly immersed in the learning process.

This is particularly valuable for three groups.

First: professionals who need to quickly familiarize themselves with related topics. A controller who wants to perform ad-hoc data analysis with Python for the first time. A learning professional encountering a new topic. Until now, that meant, search for a course, find a date, and wait. With Vibe Learning, it’s: ask a question, get started, and see initial results in 20 minutes.

Second: organizations that need broad-based upskilling. When thousands of employees need to build new skills—whether in AI, sustainability, or data analysis—the traditional training approach reaches its limits. Too slow, too expensive, not personalized enough. Vibe Learning doesn’t replace structured programs, but it enables a first, anxiety-free encounter with the topic. After all, who likes to ask a beginner’s question in a seminar and risk looking foolish by admitting they don’t know something? Often, this very first step is the biggest hurdle.

Third: And this is something that’s easy to overlook. Everyone who feels hesitant about AI will benefit when Vibe Learning is actively incorporated into a program. That’s because every Vibe Learning session is also a practical exercise in AI Literacy. You learn to craft prompts, evaluate results critically, and assess the tool’s limitations—not through a course on AI, but through hands-on use. Anyone who has experienced that AI isn’t always right and that a well-phrased question delivers better results than a vague one has learned more about AI than in many e-learning courses. "Literacy by doing" instead of "death by PowerPoint".

How can we prevent hallucinations and misinformation?

Jan Foelsing: The honest answer is, we can’t. And recognizing this fact is the first step toward dealing with it effectively.

Perhaps we should first take a moment to clarify what hallucinations actually are in the context of AI: AI systems generate answers based on statistical probabilities, not through scientific research. Even when they are factually incorrect, they can sound very convincing: they "hallucinate," as it’s called in technical terms. This is particularly tricky with topics that are new to the learner. Without prior knowledge, there’s no internal reference point to recognize a wrong answer as such. How am I supposed to evaluate an answer if I have no prior knowledge of the topic?

The solution is not better technology—even though the models are constantly improving—but rather how Vibe Learning is meaningfully integrated into the learning process. The most important factor is social validation following AI exploration. Our experiments have consistently shown that gaps and errors only became apparent once learners had to explain their results to others. What sounded coherent in the AI dialogue sometimes crumbled under the scrutiny of peers. This step is not optional; it is the actual quality filter and is also essential for enabling deeper learning.

Furthermore, it should be noted that a topic should be chosen in which the models used are expected to provide high-quality answers. A standard tool, for example, will not be able to address company-specific topics. When setting the framework for Vibe Learning, learners should also be made aware of potential risks, and tools for critical thinking and fact-checking should be made available.

Does this type of learning also anchor the learning content, or is it more a case of "learning for the moment"?

Jan Foelsing: That’s a very important question, and the answer again depends on how you use Vibe Learning.

Simply using AI dialogue—for example, 20 minutes of exploring a new topic—primarily generates quick "aha" moments. These feel genuine and satisfying but are often fragile. Learning research refers to this phenomenon as the "fluency illusion": When something is presented in an easily understandable way, it creates the feeling that you’ve understood it—even if, upon closer inspection, that understanding remains superficial. AI amplifies this effect because it’s optimized to provide understandable and affirming answers. It rarely contradicts or challenges you.

For this reason, Vibe Learning is explicitly an entry point for us, not a substitute for in-depth learning. The learning process is not anchored in the exploration itself, but in what comes after. Therefore, we work with a four-part approach: 1. Ignition – sparking enthusiasm with a good question. 2. Enablement – clarifying the framework, briefly introducing tools and strategies. 3. Exploration – the actual self-directed AI learning phase. And 4. Harvesting – the collective harvest: sharing, comparing, discussing, and defending results.

A real-life example: During a session at the New Learning Lab, participants used an AI-powered dialogue to explore the concept of entropy. In the exploration phase, everyone developed their own analogies, such as: "Entropy in organizations is like an office without a filing system: it still works for a while, but the clutter keeps piling up." During the harvesting phase, it became clear that some analogies didn’t hold up upon closer inspection. This corrective action by the group—not by the AI—was the actual learning moment.

In addition, we always include a fifth step: Next Steps. How does the initial impulse become a specific activity—a learning circle, a work project, or an in-depth exploration in a peer group format? Even the best session remains an isolated experience without this transfer. The architecture determines whether Vibe Learning remains a momentary experience or becomes the starting point of a sustainable learning process.

For which types of content is this learning method suitable, and for which not so much?

MJan Foelsing: Vibe Learning works best with exploratory topics. Basically, it’s ideal whenever the goal is to understand a new topic, identify connections, and form initial mental models. Conceptual topics such as agility, innovation, sustainability, or systems thinking are ideal because there isn’t a single "right" answer. Instead, there is a space of understanding that can be explored. Interdisciplinary questions also benefit because AI can effortlessly bridge gaps between disciplines that a single expert rarely covers.

It is also well-suited as an introduction to technical topics that previously seemed overwhelming. Someone who has never worked with data analysis can understand in half an hour what a data set is, how a pivot table works, and what questions can be asked of data—without writing a single line of code. Getting started becomes an experience rather than a hindrance.

Vibe Learning is less suitable for content where precision is more important than exploration. Compliance training, safety-critical procedures, and regulatory requirements with zero tolerance for errors—in these cases, exploratory AI dialogue is risky because hallucinations are not just a learning issue but can have real-world consequences.

When it comes to topics that rely strongly on exact figures, legal paragraphs, or process steps—such as tax law or chemical processes—other formats with verified source material are needed. The trick is to make a conscious decision: Where does the exploratory approach help spark curiosity and understanding? And where is the reliability of verified content needed? Both have their place—it’s L&D’s job to orchestrate this effectively.

How much experience has been gained so far?

Jan Foelsing: We’ve gained the most in-depth experience in the New Learning Lab—a closed community for learning professionals where we regularly test new learning formats. The most impactful experiment was a session in which participants were asked to explore the concept of entropy through an AI dialogue in 20 minutes and apply it to their professional context—a topic most of them were unfamiliar with beforehand.

The diversity of the results was the most remarkable aspect. Participants used a wide variety of tools—ChatGPT, Claude, NotebookLM, meinGPT—and completely different strategies. While some sought explanations, others had the AI ask specific questions, and others developed analogies or let the AI guide them through the topic via voice dialogue. One participant even generated a song on the topic using the AI music platform Suno. The spectrum ranged from physically accurate definitions to creative interpretations such as "Entropy in organizations is the creeping loss of order when no one actively maintains structures."

Initial ideas for integrating this into existing learning programs and practical implementation have begun. However, what we don’t yet have is reliable data on effectiveness. To what extent does the content actually stick with learners? How does the learning effect differ from traditional formats? How do we measure transfer? These are open questions that will keep us busy in the coming months. Vibe Learning is just emerging from the experimental phase—and right now, systematic testing and honest evaluation are needed instead of just enthusiasm.