Hacker News has a hidden superpower, and it isn't the front page. It's the comments.
That's where people quietly say "you have to watch this", "read this book", "just use this tool". Then the thread sinks and the advice gets buried.
This week, three projects dug it back up.
📺 First: HackerTube
@karpathy built HackerTube: the videos HN commenters recommend to each other.
I liked it so much I turned the idea into a channel on DEV·TV, my retro TV for developers. Channel 15 HN PICKS now plays HN-recommended videos, each showing how many users recommended it.
📚 Then: HackerBooks
A day later, HackerBooks hit the HN front page: ~16.5k books that HN commenters brought up, from 2007 till now.
Same idea, new shelf.
🛠️ Now: HackerTools
Videos and books were covered. But the question developers ask HN most is "what tool should I use?"
So I built HackerTools, soon to be part of DEV·TV.
It ranks ~116 developer tools by how many different HN commenters recommended them: editors, terminals, git clients, databases, infra, self-hosted apps and productivity tools. Every tool links back to the real comments, so you can read why people love it.
The problem with counting mentions
The obvious approach is to count how often a tool's name appears. But mentions are noisy:
- "I switched to Neovim and never looked back" ✅
- "Neovim crashed on me again" ❌
- "Alfred Bester was his favourite role" ❌ (a real false match from my first run)
So HackerTools only counts a comment when a recommendation phrase sits close to the tool name:
STRONG = re.compile(
r"\b(i use|been using|switched to|highly recommend|"
r"daily driver|my go-?to|can't live without|swear by)\b", re.I)
WINDOW = 120 # characters either side of the tool name
It also:
- counts people, not comments, so one person praising tmux 20 times is one vote
- skips job posts, since "Who is hiring" threads list stacks, not recommendations
- matches ambiguous names strictly: Zed, Warp and Bun are case-sensitive, and Alfred only counts as "Alfred app" or "Alfred workflow"
- picks the best quote for each tool, scoring first-person comments with a close cue higher
It's still a signal, not a verdict. But it reads like recommendations, not noise.
Top comments (0)