Don't build a spaced repetition startup (www.giacomoran.com)

Giacomo Ran and co‑founder Andrea Vaccari are pausing full‑time work on Rember, their AI‑assisted spaced‑repetition app that auto‑generates flashcards from what you read. Although the product was clean, opinionated, and loved by its creator, it failed to find venture‑scale traction. The writeup argues that general‑purpose spaced repetition faces three structural frictions that limit growth: durable learning requires “desirable difficulties” (effort), habit formation is hard (daily reviews compete with social apps), and self‑directed learning forces continual decisions about what to remember. Popular incumbents (Anki, Duolingo) and tutors sidestep some frictions with shared decks, fixed curricula, or gamified progression — advantages a general tool lacks. For the AI/ML community the post is a cautionary case study: large language models largely solve “construction” (turning ideas into cards) but not “targeting” (deciding what’s worth remembering). That means key technical challenges remain — modeling user goals and prior knowledge, deduplicating across sources, sequencing prerequisites, and handling brittle LLM outputs — plus UX tradeoffs between invisible automation and user control. Prompt engineering and human‑in‑the‑loop review mitigate errors but reintroduce friction. The takeaway: AI lowers card‑creation cost but doesn’t remove the behavioral and product design barriers that determine adoption; startups in this space must plan explicitly to absorb, redirect, or remove those frictions or focus on narrower, curriculum‑driven niches. Rember will be maintained as a side project.