Year in Review: What 2026 Taught Me About Reading Documentation

I started 2026 with a goal.

Read more docs. Less Twitter.

I sort of pulled it off.

I read 23 documents end-to-end and used another 18 as references. I also still spent more time on Twitter than I'd like to admit. Progress, not perfection.

Here's what 2026 actually taught me.

Table of Contents

• Lessons Learned
  • 1. Generation Is the Commitment Moment
  • 2. The Right Medium for the Right Kind of Reading
  • 3. Pinning > Searching
  • 4. AI Assistants Don't Replace Reading
  • 5. The Saturday Morning Habit Compounds
• What Changed About My Reading Workflow
• The Thing I Got Wrong
• The Other Thing I Got Wrong
• 2027 Plans
• The Thing Nobody Talks About
• Set a 2027 Reading Goal

Lessons Learned

1. Generation Is the Commitment Moment

I noticed this halfway through the year.

When a docs site is just a tab, my completion rate is ~10%. When it's a PDF I generated, it's ~50%.

The act of generation is small (~2 minutes) but it changes my relationship with the content. I made a deliberate choice. The PDF lives on my drive next to my projects. It's mine in a way the tab never is.

If you're trying to build a reading habit, the act of generating > the act of bookmarking > the act of leaving a tab open. Generate.

2. The Right Medium for the Right Kind of Reading

I used to read everything in a browser. The browser was the universal reader.

In 2026, I split:

• Browser: quick lookups, "remind me what this method takes"
• PDF on tablet: sustained learning, weekend reading, conceptual chapters
• Printed: the very best stuff (Rust book, Stripe webhooks page). Annotated, dog-eared, on the desk.

That's three media for three jobs. The browser was doing all three poorly. Now each medium does its one job well.

3. Pinning > Searching

When I needed an old reference (the Anthropic SDK docs from six months ago, the OpenAI behavior from a year ago), I'd hit a wall on the live docs site.

The live docs always show now. They can't show "what this looked like in March 2026" unless I have a frozen reference.

A pinned PDF is that frozen reference. The naming convention I settled on: <library>-<api-version>-<sdk-version>-<date>.pdf. Verbose. Worth it.

This is the same pattern that made the Anthropic API workflow, Stripe, and OpenAI so valuable in 2026.

4. AI Assistants Don't Replace Reading

I had a stretch in March where I let Claude / GPT answer most of my questions instead of reading docs. It worked. For 60% of questions.

The other 40% — the deep "why does this work this way" questions, the "what was this design decision" archaeology, the "I need to understand this whole subsystem" requests — needed me to read.

AI assistance is a fantastic complement to reading. It's a poor substitute.

I now have a rule: if a topic matters for ongoing work, I read the docs first, then use the AI assistant for tactical lookups. The order matters.

5. The Saturday Morning Habit Compounds

The single highest-leverage change in 2026: Saturday morning, 7-9am, with one PDF on my Boox.

90 minutes of focused reading per week. ~50 hours over the year. ~3,000 pages.

That's a graduate-school-level reading volume on top of normal work. It compounded. By December, I was noticeably ahead of January-me on three different topics.

For more on the e-ink workflow, the e-ink devices guide covers the setup.

What Changed About My Reading Workflow

January-me to December-me, side-by-side:

That's a real workflow shift, not a marginal one. The PDF habit was the keystone.

The Thing I Got Wrong

I bought too many devices.

I started 2026 with a Boox. Bought a Kindle Scribe in April thinking I'd use it more for long-form. Ended up using it 10% as much as the Boox.

The Boox would have been enough. I don't regret the Scribe (it's a beautiful device for writing on PDFs), but I'd have done fine with one device, not two.

Lesson: buy the device, use the device, don't buy the second device until you've genuinely outgrown the first.

I also got the e-ink screen size wrong on first purchase. The 7-inch Boox Page felt great in the store; in practice, it was too small for code-heavy PDFs. The 10-inch Tab Ultra C is what I actually wanted.

If you're new to e-ink: 10 inches, color, before any other consideration.

The Other Thing I Got Wrong

I generated too many "aspirational" PDFs.

47 generated, 6 never opened. Those 6 represent topics I thought I'd care about and didn't. Not a big deal — generation is cheap — but it's a small flag that I was sometimes generating as a substitute for reading.

In 2027, I'll be more honest about what I'll actually finish. Generate less, read more.

2027 Plans

The simple plan:

One big PDF per month. A book-shaped doc, read end-to-end. December 31, 2027 — 12 deeply-read documents. 12 topics I'll actually understand.

Saturday morning ritual continues. It's working. Don't fix it.

More printed material. The two PDFs I printed in 2026 (Rust book, Stripe webhooks) became the most-referenced things on my desk. Print 4-5 more in 2027.

Less aspirational generation. Pick the PDFs I'll actually finish. Generate them. Skip the "I might need this someday" ones.

That's the whole plan. Ambition: read 12 things deeply. The rest is execution.

The Thing Nobody Talks About

A reading habit isn't only about productivity.

There's something quietly different about a Saturday morning where you sit with a PDF and a coffee and don't open Twitter for two hours.

The thinking is different. The mood is different. Your relationship to "knowing things" shifts from "I can search for this" to "I have read this and remember enough to ship."

The first version is fine. The second version is what makes engineers feel like engineers.

2026 was the year I noticed the difference.

Set a 2027 Reading Goal

offlinedocs.ai/new → generate the first PDF you want to read in 2027.

One PDF. Today. The first one.

January will be too busy. Generate it now.

12 months from now, December 28, 2027, you'll write your own version of this article.

What will it say? Let's find out.