My 2026 PDF Bookshelf: Every Doc I Actually Read
It's that time of year.
Table of Contents
- The 2026 Reading Audit
- Per-Month: What I Generated vs Read
- The Surprises
- The 2027 Starter Shelf
- What I Got Wrong
- What Surprised Me Most
- Audit Your Year
The 2026 Reading Audit
I tracked every PDF I generated and how many of them I actually finished.
Generated: 47 PDFs. Finished (read all the way through): 23. Used as a reference (read 30%+, kept on hand): 18. Generated and never opened: 6.
A 49% finish rate is honestly high for me. The 6 I never opened: I generated them in a moment of optimism and never had a reason to come back. That's fine. Generation is cheap. Two minutes per PDF.
The signal in this data: the act of generating a PDF is what makes me commit to it. The browser tab equivalent is closer to 10% completion.
Per-Month: What I Generated vs Read
January
- Generated: Postgres tutorial, MDN HTTP, Tailwind v4 docs
- Read: Postgres tutorial cover-to-cover, MDN HTTP about half
- Skipped: Tailwind v4 (used as reference only)
February
- Generated: TanStack Query, TanStack Router, the Vercel AI SDK
- Read: All three, mostly in cohort with side projects
March
- Generated: Rust Book (printed!)
- Read: Cover-to-cover. Took ~7 weeks. Worth every page.
April
- Generated: Anthropic API, OpenAI API
- Read: API references, several times for specific features
May
- Generated: Laravel docs, Postgres full manual
- Read: Laravel for a side project, Postgres for performance work
June
- Generated: Svelte handbook, TanStack Table
- Read: TanStack Table twice (the why-I-did-it-twice story)
July
- Generated: Next.js, Vercel AI SDK v5 (re-pin)
- Read: Next.js App Router fully, AI SDK as reference
August
- Generated: Python stdlib (focused), Rust std (kept as reference)
- Read: Python
itertools/functools/pathlibend-to-end
September
- Generated: Angular (didn't end up using), Stripe API
- Read: Stripe API for a payments project
October
- Generated: Confluence space (work), Anthropic API v2 (re-pin)
- Read: Confluence (selectively), Anthropic for prompt caching specifically
November
- Generated: Various conference prep PDFs
- Read: Most before the conference
December
- Generated: This year's reading audit material
- Read: This article's research
The Surprises
Docs I Didn't Expect to Love
Postgres full manual. I generated it expecting to skim. Ended up reading the WAL chapters thoroughly because something at work needed it. The Postgres docs reward depth.
The Anthropic developer guides. I bundled these mostly as a "reference, just in case" PDF. Ended up using the prompt caching guide three times. Saved a lot of money.
MDN HTTP. A bootcamp grad I mentor was struggling with HTTP fundamentals. I re-read the MDN HTTP docs for the first time in 5 years to teach. Discovered a few things I'd been doing slightly wrong. Bootcamps don't teach this stuff well; the MDN docs do.
Docs I Expected to Love and Didn't
Angular. Generated, opened, read 30 pages, didn't end up using Angular this year. The PDF sits on my drive. Maybe 2027.
Bun docs. Started reading. Got distracted. Ended up using Bun in production but learned by doing, not by reading. Filed the Bun PDF for "later, more carefully."
Various AI infrastructure tools. I bundled docs for ~5 tools I thought I'd use. I used one. The other four PDFs are unread.
This is normal. Generation is aspirational; reading is reality. The ratio of "wanted to read" to "read" is what it is.
The 2027 Starter Shelf
What I'd recommend to someone who wants to do this for 2027:
| Why | |
|---|---|
| Your primary backend language's stdlib | Daily reference |
| Your primary frontend framework's handbook | Conceptual fluency |
| Postgres tutorial (or your DB equivalent) | Most apps need this |
| Your primary AI SDK reference | If you build AI features |
| One "deep-cut" book per quarter | Rust book, SICP, designing data-intensive applications, etc. |
| Your team's internal docs | Required reading |
That's 6-8 PDFs. Maybe 1,500 pages of technical reading over a year. Doable at 30 minutes a day, less time than the average person spends scrolling.
For the system around this, the personal dev library guide walks through the longer-term pattern.
What I Got Wrong
Three things I'd do differently:
1. I generated too many "aspirational" PDFs. A 49% finish rate is good but the real metric is "did I read what I planned to read?" — and there I was closer to 60%. Be honest about what you'll actually read before you generate.
2. I underestimated the value of pinning. I should have pinned my Anthropic and OpenAI PDFs more often. I ended up wishing I had a v4.5 reference for some old code.
3. I should have printed more. I printed two PDFs in 2026 (Rust book, the Stripe webhooks page). Both turned out to be the most-referenced technical reading I did. Print more in 2027.
For the longer print argument, the print guide covers it.
What Surprised Me Most
The number of times I closed my browser entirely while reading.
In 2025, my browser was always open. Documentation lived there. Tabs accumulated.
In 2026, I noticed myself closing the browser, opening a PDF, and reading for 90 minutes without re-opening it. That's a specific behavior change. It maps directly to mental focus.
The act of generating a PDF is small. The act of reading it without a browser is, somehow, large.
Audit Your Year
offlinedocs.ai/new → generate a "best of 2026" PDF for your own bookshelf.
Pick the 3-5 docs you actually used most in 2026. Bundle them as a single "year in review" PDF. Save it.
Future-you, looking back at 2026, will appreciate the snapshot.
That's a 5-minute project that pays off for years.
The PDF bookshelf isn't only about productivity. It's about deliberateness — choosing what to read, instead of being chosen for you by an algorithm. 2026 was the year I noticed the difference. 2027 will be the year I lean further into it.
Ready to Get Started?
Start creating your Offline Docs Now! Reduce screen time and save your eyes.