Products#Coding agent#Data analysis
HackerBooks: 15,000+ books mined from HN comment mentions, built almost entirely with Claude
HackerBooks indexes 15,362 books mentioned in HN comments since 2007; its author says Claude wrote almost all of the site.
On November 1, a developer going by roansh posted a Show HN launch for HackerBooks, a site that combs twenty years of Hacker News comments — 2007 to 2026 — for books people actually bring up and ranks them in a browsable catalogue; within four hours of posting it had collected 467 points and 123 comments (Internet Archive snapshot, 19:56 UTC that day). According to the site’s own public data file (generated November 2), it holds 15,362 books — the launch post says roughly 16.5k — each aligned with Open Library metadata and linked back to the comments that mention it. The author says Claude (Opus 5.5) wrote almost the whole thing, and that his own job was mostly steering and making calls.
Key points
- The 15,362 books were surfaced from 270,334 mentions in 217,954 comments by 56,456 commenters, spanning 2007–2026; a book gets in only when at least two different commenters have brought it up.
- The all-time leaders: The Art of Computer Programming (1,798 mentions by 1,125 users), Gödel, Escher, Bach (1,386), Snow Crash (1,283) and Thinking, Fast and Slow (1,165).
- A mention is not an endorsement: Hitler’s Mein Kampf sits in the top 600 on 535 mentioners, and the author changed the site’s wording from “recommended by” to “mentioned by” mid-thread.
Background
The idea is a direct homage. Karpathy built HackerTube — HN comment mentions of videos, turned into a browsable site — and roansh’s launch post describes his version as “HackerTube, but for books.” We have covered this pattern before: AI letting a single developer build in days what used to take a team, as when another developer ported an entire Call of Duty map into the browser with Claude (our earlier report).
The wave is spreading in real time. The same day HackerBooks launched, a user named techscruggs posted the inevitable follow-through — “HackerBooks, but for all the posts on HN” — a catalogue of every post rather than books. It managed 2 points and zero comments. Commenters on the original thread also pointed out that scrape-HN-comments projects have surfaced repeatedly over the years and mostly go unmaintained once they fall off the front page. And despite the name, this has nothing to do with hackerbooks.com, a 351-point Show HN from 2011.
The facts
The site itself is deliberately thin infrastructure. In the comments the author describes pure JavaScript plus static JSON files on Cloudflare, with book covers served from R2; the first load pulls only a ~30 KB index of the top books while the full catalogue of 16k loads in the background, with hashed filenames so pages cache essentially forever. The pipeline is “python + duckdb over the HN dump” — no backend, no database service. Here is the shape of the data:
| Item | Number or rule | Source |
|---|---|---|
| Books indexed | 15,362 (data file); ~16.5k per launch post | Site data file, generated Nov 2, 2026 |
| Total mentions | 270,334, across 217,954 comments | Same |
| Commenters involved | 56,456 | Same |
| Year range | 2007–2026 | Same |
| Inclusion rule | At least two different commenters per book | Launch post |
| Comments stored per book | Capped at 30 | Author in comments |
Both book counts are real but not the same: the launch post’s ~16.5k was the author’s figure at submission time, while 15,362 comes from the data file regenerated on November 2 — the pipeline is still being adjusted, so we keep both rather than picking one. The data is downloadable from the About page, which is what separates this from most of its predecessors: the numbers can be checked.
Three bugs found by readers within the first 24 hours were all fixed in the thread. The “mentions since” date filter did not work until the author rewrote it to use first-mention dates; Vanity Fair the magazine cracked the book rankings until the author counted only comments naming Thackeray; and a broken “show more comments” button got a same-day fix.
What others say
The thread’s criticism landed early and the author took most of it head-on. User jchook liked the site but objected to the launch post’s prose: “Please don’t put generated text in HN posts.” The author’s reply: he had Claude draft it, edited it only twice, and should have written it himself.
The sharpest hit was the mention-counting flaw. User drayfield flagged that Mein Kampf had been labeled “recommended by 651 HNers”; the author changed the wording to “mentioned by” and promised more filtering work. Per the November 2 data file the book still sits high, with 625 mentions by 535 users — on the same page as The Communist Manifesto at 301 mentions. User danso pressed the deeper question: if Claude built everything, who does the judgment calls on filtering noise? The author’s answer was “coming right along.” Others skipped the nuances entirely — users like Aurornis noted this kind of site has been built many times, and usually dies with its front-page traffic.
Our take
For anyone who already lives on HN, HackerBooks is a ready-made consensus reading list — sortable by mentions, viewable by year, with every number pinned back to its source comment. For people building data products, it doubles as a demo of how far duckdb plus static files can take a 270k-mention community dataset. But the site’s own caveat is the honest one: a mention counts the same whether a commenter loved the book or cited it as a disaster, so the top of the list mixes canon with Mein Kampf, and both the pipeline and the filters are still moving — that is why the two book counts disagree. As for “Claude wrote almost all of it,” the thread argued about it all day without resolving anything. The author says even the data pipeline was Claude’s work; the danso line of questioning amounts to: who is accountable when it’s wrong? Judging by the three same-day fixes, the answer is still the human. Whether the whole pattern survives its fifteen minutes is checkable in a month — see if the data file keeps regenerating.
How to try it
Hackerbooks.org needs no account: the home page is a reshuffled daily shelf, and the top nav leads to Top all-time, Trends by year, and Rising for recent climbers (the shortcuts are g then h/t/m/r; press ? for the full list). On each book page every mention links back to the original comment, so you can read the context. To work with the raw material, download the book-level CSV dataset from the About page (data/hackerbooks.csv) and open it in DuckDB or any spreadsheet tool. The full argument is on the Show HN thread.