Chris’ Corner: What have you learned?
One of my favorite kinds of writing is someone sharing something they are knowledgable about because they actively learned something over the years they have been doing it. Here’s some of those articles.
The “build vs buy” conversion in business is fascinating to me. It feels like you should always be able to analyze it and come up with the right answer. It feels like you should be able to develop a system to make these decisions that feels right for your company. Like “we have a lot of talented engineers and value control, so we lean toward building unless it’s too tangential to our core business”. Or “we’ve just taken a big funding round and haven’t fully staffed up yet, so we lean toward buying so we can keep moving fast.” But it’s all contingent on price, features, support, integrations, etc.
Ted Neward weighs in on this with an important point: you have have a certain amount of context when you make the choice, but that context changes over time and is different when you evaluate that choice.
You’re never going to be always right. Or, to be more accurate, the decision you make today has absolutely no guarantee that it’s going to be the right decision the further away we get from the decision point. The best we can do is make the decision for today, accept that we pay an opportunity cost in doing so, and keep an eye for when the context has changed enough such that we need to consider revisiting the decision.
It feels like a rite of passage when you decide that boring software is good software. It’s (probably) boring because it works, has worked for a long time, and doesn’t have flashy releases. That means it never makes the news and tends not to be overly celebrated, but it probably should be. I certainly had my moment. I wouldn’t touch a released-this-week framework with a 10-foot pole for a serious production project.
Scott Jenson is looking at the world of LLMs as possibly a technology that should get more boring. I’m certainly in the majority view here as well with this stuff. Use LLMs/AI for what they are good at… only. We’re going to get there eventually anyway.
Ultimately, a mature technology doesn’t look like magic; it looks like infrastructure. It gets smaller, more reliable, and much more boring.
Speaking of AI stuff, maybe it’s just my bubble, but I don’t see that many articles of people talking about how they use AI and what for. My guess is that people either don’t want to be thought of as having an AI crutch already mixed with being worried if they write about it, they’d be showcasing their own ignorance by not using it as well others.
David Bauer doesn’t seem to be detered though, he’s got a whole How I Use AI page with just enough detail.
Just to get over my own hangups: my own usage is 95% coding work, mostly through Cursor (although I’m ready to give Zed + Claude Code a proper shot) and mostly agent-style, where I take my time crafting prompts that explain exactly what I want and the files I know are involved and let it have a crack at things. I find it pretty damn helpful, honestly, but I happen to be doing work lately that is a lot of slight behavior adjustments to code in a well-established code base. I think AI is much better at that than from-scratch situations or where it has too much freedom with architecture choices.
Eric Bailey knows a hell of a lot about making accessible websites. ARIA is an inevitable topic in that world. Not the usage of ARIA, mind you, as accessibility practitioners are quick to tell you that ideally you don’t need it at all. But you should let Eric tell you: What I Wish Someone Told Me When I Was Getting Into ARIA.
ARIA came up, believe it or not, sigh, in the AI scene last week as well. OpenAI released ChatGPT Atlas, the latest “AI browser”. Part of me is impressed. The first thing I did was try “Agent Mode” and type “go to codepen and make a css rainbow” and sure as shit it immediately navigated to the CodePen editor and got it done. It put the CSS in the HTML editor at first, but understood the mistake and fixed it. I like the idea of “tell computer what to do and computer does it”.
But there are serious concerns with it, going as far as saying it’s a browser that is straight up anti-web. I get the concern: it doesn’t want to link you away to websites, it wants to just give you the information you’re likely looking for. That’s been an issue with AI since the beginning. What happens to the web when the giants of the current web are actively trying to keep you away from websites? OpenAI would never say that’s what they are doing, but c’mon. Instead, they say, help us make your website more ChatGPT friendly! Sprinkle it with ARIA attributes!!!
Lemme just leave you with a couple more.
- Sean Goedecke: What is “good taste” in software engineering? — “… immature engineers are too inflexible about their taste. They know what they like, but they mistake that liking for a principled engineering position.
- Andrey Sitnik: What we learned from creating PostCSS — Interesting that they think the all-plugins approach went too far (“people don’t like to make choices, and most projects need the same features anyway.”)
