Don't Compete With Existing EdTech. Make It Obsolete.

01 /The Concept

Rabbithole's creators describe it as "a launchpad for the curious" – which is a reasonable summary of what happens when you use it.

You start by typing a question. The AI returns a concise answer – but the real feature is what comes alongside it: three pre-formulated follow-up questions that naturally branch from the answer you just received. You can take one of those, or ask your own.

Each branch generates another answer plus another trio of suggested questions. Over time the interface builds an expanding tree of interconnected questions and answers, navigable in any direction. You can also backtrack to earlier nodes and follow paths you didn't take the first time, spinning up new branches wherever curiosity leads.

All previous sessions are saved, so at any point you can return to an old thread, inspect the tree you built, and continue exploring it in new directions by clicking on unexplored nodes.

Rabbithole surfaced on Product Hunt the day of its launch.

02 /Why It Matters

Rabbithole, in its current state, isn't a fully formed startup with a revenue model and a defined audience. It's a proof of concept – a first experiment in a new interaction paradigm. More specifically, it's a UI concept that could be transplanted into quite different products.

The concept landed personally, partly for historical reasons. There was a moment managing a search product that was fading fast – a moment that demanded finding genuinely differentiated approaches, not just incremental improvements over whatever the dominant player was doing. One idea from that era was a search results page you could navigate "sideways" – clicking a small icon next to any result would automatically refine the query to explore the specific sub-topic that result touched on.

For a query like "Sony headphones," the standard results page is necessarily vague – mixing pricing, reviews, forum discussions, troubleshooting. Clicking sideways next to a review link would surface more review content. Clicking next to a forum link would surface community discussions. From each of those new pages, you could keep shifting sideways, drilling as deep and as specifically as your interest carried you. The result: not a linear list of results but a navigable tree of possibilities.

A commenter on Product Hunt described Rabbithole as a "decision tree that builds itself dynamically" as you explore a topic. One of the founders confirmed this framing was resonating with early users and that they planned to lean into it. Other commenters noted they already approximate the same thing manually with ChatGPT – firing follow-up questions off each answer, building an implicit tree in their heads. Rabbithole just makes that tree visible and navigable.

03 /Opportunities

The most natural application for this kind of non-linear exploration is in education platforms – most of which are built on strictly linear architectures, from course structures (ordered lessons) down to the lessons themselves (ordered content blocks).

There are exceptions. Kyron Learning ([covered here](/review/teper-vygodnee-ispravljat-chem-objasnjat)), backed by $20M, built what it calls a "truly interactive" children's learning platform. Courses remain sequential, but lessons branch: after each content chunk, a student answers comprehension questions. Correct answers advance the sequence; incorrect ones route to an alternative explanation targeting the specific misunderstanding, followed by another check. The path through the lesson isn't predetermined – it adapts to where the student's understanding breaks down.

The underlying insight is striking: misunderstanding drives learning. If a student already understands everything, there's nothing to teach. The teacher's job is to anticipate the wrong mental models students will carry in and prepare explanations that correct each one. The AI's job is to match student responses to likely misunderstandings and serve the most relevant fix.

The next step – having AI generate the misunderstanding map and the correction library, then refine them through real-world usage – brings the concept very close to what Rabbithole's AI does when it builds its question tree. Both are dynamic structures that deepen based on where the user gets lost.

The broader opportunity is adaptive learning platforms with two core properties:

- Automatic adjustment to each learner's knowledge level and specific gaps - Open-ended paths that let learners explore material in whichever direction genuinely interests them

Neither of those properties can be designed manually at scale. AI is the enabling layer – which makes this a timely place to build.

Obrizium ([related review](/review/prodavat-znanija-a-ne-kursy)), which raised £12.4M, has been calling this "truly adaptive" learning (the label "non-linear" was its earlier framing). It operates in corporate training and is explicitly focused on closing individual knowledge gaps rather than delivering uniform curricula.

Education is a permanent, enormous market. Most existing platforms teach in the old linear way. That's the opening – and the architectural advantages of AI-native, non-linear alternatives grow more visible by the year.