Do you know what needs an agentic layer? Car dealerships.

Why am I getting porn in my bluesky reels. Fuck this. This is why I quit X.

Thoughts regarding code health or quality success metric. If you can't write an LLM prompt for it and have Claude reliably produce a percentage, it's probably a bad success metric. This might even extend to stuff outside code quality with various MCPs at our disposal.

Why is zuck's company announcement about Meta Compute just permanently pinned on top of my Facebook feed. Its obnoxious.

Someone needs to create an AI chat version of "let me google that for you".

I set my iphone to greyscale and set my triple back tap to toggle it. Hoping this will combat screen time and phone addiction. Feeling optimistic so far.

It’s just infuriating how bad we are at error handling and error UI. It’s just a trade off we’ve seemed to have accepted as an industry even though we have the patterns to fix this (e.g. error-as-data + exhaustive typing as in some FP languages).

Every once in a while I pop back in to visit my alma mater pinterest.com and I'm impressed by how fast the site loads and feels, especially recently and compared to when I used to worked there. Sometimes you forget there are really impressive engineers working at these big companies.

Because LLMs have made knowledge accessible, I find us making better decisions now. People can't just bluff their way through making technical decisions anymore since everyone has a know-it-all at their fingertips. I think this is where experience really separates you.

Wow fuck this guy Yamamoto 😭

Claude Code is basically TDD as a service.

"It's flaky' often used as a rebuttal to integration tests. But I think flakiness is the point. Maybe this feature is just unstable or indeterministic and you're choosing to ignore it. Granted testing async flows or time-based interactions can be hard, but flakiness is more often than not an excuse.

I also think this unit testing dogma is a symptom of the pendulum swing to dynamic programming languages. With it swinging back to static type checking, you can get very far with no unit tests.

I think people get duped by ease and volume of writing small isolated tests. You point to the "test pyramid". You feel good about Best Practices™. It's like the junior engineer comfort blanket. But the fastest way (IMO) to demonstrate what you've built works is to go straight to integration tests.

I always get so much snark whenever I suggest integration testing. But it's either putting your head in your sand and pretending your product works, or doing the hard work of maintaining a test bed and treating each failure, even flake, as "a customer is experiencing this".

Uggh I made the mistake of randomly deciding to open X and that photo of the Vercel CEO taking a selfie next to a literal war criminal is living rent free in my head. Lack of empathy some people have I cannot understand…

I tend to want to work with 1 more. They tend to make better decisions. Don't cargo cult. Are not afraid of digging into code. 2 might lead to faster promotion and an identity as someone who "gets shit done" but I think you hit a ceiling.

There's a continuum of engineers:
1. One who puts there head down and struggles through problems. Uses first principles and sometimes asks for help.
2. One who searches for solutions: hyperactive brain, using AI agents, pattern matching, doesn't read error message, defaults to outside help, etc.

Trying to figure out if there is a way to give the VSCode AI your database schema so that it can better autocomplete SQL queries.

Maybe there are parallels between the 'iPad kid" phenomenon and a coder reliant on coding agents. I wonder if a coder's ability for text motion/nav/manip trick shots is going to be a thing of past. We'll have a generation of "Cursor kids" who are amused by boomer vim combos and bash-fu.

The reality though is there are things eng can do and imagine that non-eng functions cannot. DX is a classical example where it often suffers. But technology advances from research is another (local first, react, gmail, etc.). Eng sometimes needs space to broaden the solution space for product.

Engineers classically struggle to communicate the value of their work due to differing levels of abstraction from management. I've found that when managers used to be eng ICs, the translation cost is lower, and empathy is higher, creating a sweet spot for collaboration/stress.

Not saying eng is good and product is bad. But its just this natural divergence of perspective on whats valuable is what causes these micro boom/bust cycles as there's a push/pull. But I'm worried that with this mode of operation thats so common everywhere, we create a lot of burn out.

There's a micro boom/bust cycle in tech companies. Non-eng functions push for a roadmap. Aggressively prioritizes. Eng loses out. Its only until eng breaks down (quitting, pleas, operation hell) that eng vision is prioritized. Value of eng work to end-users is eventually questioned, cycle repeats.

Excalidraw is a gift to humanity. It's the only whiteboarding tool I've used that gets out of the way and lets you architect stuff at the speed of thought. FigJam, Jamboard, Miro, etc. I tried them all and they don't compare. It's also local-first, has multiplayer, and its free! Like wat.

Low key the best use of AI is trying to get a Bluesky post under 300 chars.

I watched Perfect Days, about a toilet cleaner content with life. The director suggests it's about the 'common good'. But what's wild is that a Tokyo toilet cleaner can afford a big home in the city. I'd love to see a version set in SF/NYC where 'common good' is neglected due to a 2-hour commute.

Something I realized: I dislike mentee relationships where I give "career advice". I'd much rather someone come to me with a problem they're blocked on and we learn from it. Me opining about how to "get to the next level" feels like I could easily give advice that only works for me.

I think maybe I prefer using AI for an exploration of the solution space, rather than trying to find the solution itself. But I'm not sure yet. I'm worried we will raise generation of engineers with low resilience. Grinding through error messages and debugging hard problems has been formative for me

You know that shape sorter toy where babies would mindlessly bang shapes into holes until they fit? I feel like thats a lot of engineers right now when faced with an error message and equipped with AI. It doesn't feel like path to improving yourself as an engineer if all you care is the outcome.