HackerTube: build from a problem you understand

Begin with something you want to find easier

Choose one thing you repeatedly find awkward, then describe the change that would help. You might want a clearer way to find the learning material you already saved. Start with a small collection whose contents you understand.

On October 31, Andrej Karpathy described building HackerTube with Claude Code, an AI tool that can work on software projects. The site organises YouTube videos recommended in Hacker News, an online discussion community. It offers a different starting point for browsing that collection.

Karpathy reports a catalogue of 6,454 videos drawn from recommendations in comments stretching back to 2007. Those figures describe his project, not a promise that its selections will suit every reader.

Read the announcement: Andrej Karpathy: HackerTube and the build process.

Look at the work behind the finished page

Karpathy says AI helped filter the comments and build the website. He also describes setting up access to YouTube information, starting a computer to run the site and buying a domain, the address people use to reach the site.

His account puts the build at roughly two hours. That is his report of one project, not a timetable for someone learning to build. Asking AI to secure a website also does not establish that an independent security review took place.

The useful lesson is the starting point: a clear frustration, a defined collection and a result he wanted to use. You can practise that approach at a scale you can review.

Make your first version small enough to inspect

For your own exercise, use a collection you own or have permission to use. A handful of approved links is enough to sketch a page that lets someone search, filter and open useful material. This is our suggested practice task, rather than a description of new HackerTube features.

Decide what makes an entry useful before asking AI to organise it. Check whether the title matches its destination, whether two entries repeat the same thing and what happens when a link no longer works.

A community recommendation reflects that community. If you later use AI to select information for a team, ask whose needs the selection serves and which useful material it might leave out.

Practise the journey from idea to useful result

Keep the first goal visible. You are learning to explain what should improve, check the result and decide what needs another attempt.

1. Write one sentence about the problem and who experiences it.
2. Choose a small, approved collection and define what someone should be able to find.
3. Ask AI for a proposed page and try normal, missing and repeated entries.
4. Check the links and agree who keeps the collection current before sharing it.

Bring your knowledge of the problem

Our perspective is that knowing the problem is a useful contribution to an AI build. You do not need to begin with a large product idea. Bring a small example, the result you want and the questions you cannot yet answer to a training session or a conversation with a builder.