July 5th, 2026

Articles of the week

This week I picked up these articles:

Learned/Notes

Stop using conventional commits

Going in, I thought "conventional commits" meant what I already do at work: skip the fluffy "fix: blah blah blah" and instead write the actual thing that changed, like "Allow localhost:3000 in origins for server" or "Require JWT token for all operations to authorize user." Turns out no, that's not it at all. Conventional commits are the feat(api): added auth schema... style, prefixing commits with feat//fix//etc. My first reaction was "wait, they're doing what PRs already do, but in commits? Why?"

Sumner is saying that AI tools default to this format because so many open source projects use it (I tested right as I read that, and the commits Claude has been doing in my projects don't look like that), and he thinks that's a mistake and wants to push against it. The part that got me was that the scope is optional, when I agree with him that that's the most important part of the commit message.

I don't think I've ever worked somewhere with an enforced commit style like that. It's always just been <jira_id>: description, which gives you room to add the scope if you want to. I'm curious if anyone is following conventional commits? Reading the article it felt like it's a big thing in tech but... I don't feel like it is?

The 80-20 rule for your whole life

Before I started reading the article I was thinking "Is this the rule that says 80% of your outcomes come from 20% of your inputs?" and yah, it is. I've heard that that rule applies to everything, but the truth is, I don't think I've ever sat down and actually processed it, or thought about what my inputs are and if I should change them. The author applies it everywhere, not just career, and he follows these 3 principles/habits:

  • expand your knowledge
  • master your decisions
  • build and grow.

The line that I most liked: "The most valuable asset you own isn't in your brokerage account or your retirement fund. It's the mental model you carry into every decision. That model either appreciates through constant learning or depreciates through complacency. There's no neutral zone."

The common thing across all three of his buckets is deliberate exposure to things outside your core, such as reading beyond your field, studying adjacent disciplines, spending time with people who've already done what you want to do.

I spent some time processing this and just thinking after reading the article that I want to share: I do have a foundation in a couple areas — I've been doing this weekly read since December, and I've picked up genuinely good writers and blogs from it Cate Hall being one. I have a small spiritual practice too, nothing long, but consistent. But the "expand into adjacent things" part? I'm not doing that nearly enough. I started doing it a few months ago when I realized that for the past few years, I've blocked myself from looking into other things in order to remain focused on the one I thought was most important. The truth is though, I've realized how important it is to expand your knowledge and there is no way you can do that if you remain focused on just one thing. I'm curious about how prediction markets work (Polymarket, Kalshi, why Meta apparently wants in on that space too) and instead of reading/learning about it I've mostly just filed it under "seems sketchy" and moved on. That's exactly the kind of curiosity the article is telling me not to ignore.

The pulse - Impressions from visiting OpenAI, Anthropic, & Cursor

I only read the section on minimizing price-per-token at AI companies — mostly out of curiosity about whether they're doing anything other than "just use a cheaper model."

The number that actually made me be like "holy shit" was a LinkedIn post from a Senior Staff Engineer at Coinbase saying their AI spend is down nearly half. The tactics were failover routing between providers (so you're not stuck when one is down or slow), caching, and actively clearing the context window instead of letting it grow forever. None of that is anything crazy, but seeing it cut a real company's bill in half is a good sign — it means everyone else that's not doing this can do it too. Feels like there are good opportunities for engineers who specialize in infra/platform and can make a huge impact at their company.