Before recently, tech debt was the main form of time plague that tormented development projects. The ever-tempting sacrifice of a functional, but imperfect, “temporary” implementation to the dark lord in exchange for some free time now; a simple equation that has forever haunted the codebases of hobbyists and professional software engineers alike.
We’re now witnessing a new kind of debt emerge alongside the widespread adoption of AI tools for experimentation, learning, and development; a phenomenon that has recently been coined as cognitive debt.2
Where tech debt defers the cost of clean code, cognitive debt defers the cost of actual understanding. Developers can now ship projects they can’t actually build or fix, leaning on AI to fill the gaps they never close. Over time, the gap between what someone can build and what they actually know expands, leaving them dependent on increasingly expensive tools they can’t evaluate, debugging in languages they can’t read, and shipping code they can’t even begin to explain.
Software development has seen it first, but the effects of cognitive debt will only multiply as AI deepens its integration into other industries. So what is the root cause of this problem?
It’s simple: learning has not caught up to the speed and ease of building. It is too tempting for anyone curious about making something to skip straight to production, given that production can now commence with a single prompt – regardless of how good that product is actually going to be. And learning without school remains entirely self-directed. For any subject or project a user wants to learn or build, they have to find lecture material online, determine exactly what they need to cover, and build out their own curriculum of assignments and tests, otherwise the information will go in one ear and out the other. By then, most have already given up.
Massive Open Online Courses (MOOCs) valiantly attempted to solve this problem. But the cost of hand-crafting courses for every subject keeps the available topic selection slim, let alone the rather dismal dropout rates that come from rigid, untailored course structures that force the learner into a specific learning path instead of building one around their wants and needs, like a tutor would.3 The content is there; the personalization isn’t.
The answer? Learning needs to be as effortless as vibe-coding. The sheer quantity of vibe-coded software on the internet, if anything, demonstrates how curious individuals become experimenting with new subjects when technology makes creation more accessible; we now need systems that ensure they can follow their curiosity to genuine fluency. We’re calling this vibe-learning: intent-first, goal-driven instruction that meets you where you are and builds toward what you actually want to make. Instead of prompting an LLM to build an app you won't know how to edit or fix, that same query should produce a personalized learning path with a clear progression, goal-oriented checkpoints, customized assignments, and a tutor that won’t move on until you actually understand. Not a static course or a generic chatbot. A system that closes the gap between what you can build and what you actually know.
That’s what we’re building at Airistotle.