One engineering blog post a week

Why a habit rather than a burst

Knowledge decays because the field moves, not because you forget. The fix is not another course — it is a small recurring input that keeps you in contact with how good teams actually work.

One post a week is 50 posts a year. That is more real-world architecture exposure than most engineers get in a decade of tutorials, and it takes twenty minutes.

Make it survive a busy week

The habit fails for predictable reasons, so design against them:

• Attach it to something fixed. Friday afternoon, Sunday morning, the commute. A habit with no slot loses to whatever is urgent.
• One post, not a backlog. "Read more engineering blogs" produces a bookmark folder of shame. "One post, Friday" is completable.
• Read it now or delete it. A read-later list that grows is a to-do list you are failing at. Give yourself permission to drop things.
• Do not aim for comprehension of everything. Skim the parts about infrastructure you will never touch. Read closely the parts about a problem you recognise.

Where the signal is

• Cloudflare — networking, TLS, DDoS, edge. Unusually deep and unusually well written.
• Discord — real-time systems, storage migrations, scaling with a small team.
• Stripe — API design, idempotency, correctness around money. The best API writing anywhere.
• Netflix TechBlog — resilience, chaos engineering, streaming at scale.
• Figma — multiplayer, CRDTs, browser performance.
• Postgres release notes and mailing lists — dry, extremely high signal.
• Public postmortems — from anyone. Failures teach more than successes, and the good ones read like detective stories.
• Your own dependencies' changelogs — the highest relevance-per-word of anything on this list. See following your stack.

Aggregators — Hacker News, Lobsters — are for finding posts. Reading the comments is not reading the post.

The three-line notes file

One plain text file. Per post, three lines:

2026-08-14 — Discord: moving trillions of messages to ScyllaDB
Constraint: Cassandra GC pauses caused p99 latency spikes at scale
Cost: a Rust data-services layer they now maintain themselves

Thirty seconds. The value is not the archive — it is that compressing a post into three lines forces you to identify what it was actually about. A post you cannot compress is a post you skimmed.

Six months on, this file is searchable in a way your bookmarks are not, and rereading it takes five minutes.

Read for transfer

After each post, ask one question: is there anything I work on with the same constraint, at a smaller scale?

Usually no, and that is fine — you have still learned what the failure mode looks like before meeting it. Occasionally yes, and that is a concrete improvement you would not otherwise have thought of. More on this in reading engineering writeups.

Prove you know it

Read one post this week and write the three lines. Do it four weeks running. If the file has four entries at the end of the month, the habit exists; if it has one, the slot was wrong — move it.