Scrimba launches HN.watch, turning Hacker News links into AI explainer videos for about 4 cents each

Scrimba, the coding-education startup founded by Per Harald Borgen and backed by Y Combinator in 2020, has released a tool that converts written content into short explainer videos using an LLM. The company demonstrated it with HN.watch, a Hacker News reader that swaps out article links for generated video explanations, building each clip the first time a visitor clicks a post. Borgen announced the project on Hacker News, where it drew a small but immediate thread of comments.

The underlying product is called Scrimba Explain. It builds on a video format the company has used for a decade to teach programming, in which the visual output is rendered as HTML rather than generated pixel by pixel. Borgen said that approach carries visible quality tradeoffs compared with diffusion-model video, but buys three things: generation in a few seconds, a cost near $0.04 per clip, and cheap AI-assisted editing. Image generation, when a video uses it, pushes that cost up sharply.

The cost and latency figures are the pitch. Borgen argues that if producing a video drops from dollars and minutes to cents and seconds, use cases that never made economic sense start to appear. Scrimba already generates an explainer for every pull request internally, and the company points to documentation pages, course drafts, and personal projects such as wedding invitations and birthday messages for kids as near-term applications. A Chrome extension can turn an article into a video in roughly four seconds.

The stack is deliberately unusual. Scrimba's CTO, Sindre Aarsæther, created Imba, an open-source language that compiles to JavaScript and interoperates with npm and Node, and the company also built its own sync engine, OP, plus a context-management system for agents called Q. Borgen said the team worried LLMs would stumble on a stack with almost no presence in training data, but found the opposite: the density of the languages, where one declaration covers storage, sync, permissions, UI, and what the agent sees, leaves less room for a model to misread the architecture. Instead of React, Express, Supabase, and LangChain, Scrimba wrote each layer itself, which Borgen acknowledged is a case of not-invented-here syndrome.

On models, Scrimba leans on Gemini, OpenAI's GPT family, Inworld, ElevenLabs, and others for the audio and language work. The tool ships through several entry points: a web interface at scrimba.com/explain, an MCP server for coding agents, a ChatGPT plugin, and the Chrome extension, with links collected in the company's documentation.

The launch is likely to meet resistance from Hacker News readers who favor text, and Borgen said as much in his post, noting the tool is not for everyone while arguing that younger audiences increasingly prefer video. The open question is whether sub-five-cent generation holds up at scale once image generation and longer scripts enter the mix, and whether developers actually want a video layer over documentation they already skim.