azim

open source maintainers don't care that you used claude code for your PR. what actually pisses them off is when you answer their human questions with AI slop. they have their own agents, they don't need yours playing telephone.

when i pick up a new framework or tool, nothing clicks at first. i feel slow, confused, like i'm missing something obvious. it would be arrogant to expect fluency on day one. the smarter move is accepting you're the least competent person in the room, because that's the only version of yourself that actually gets better.

am i the only one who doesn't like infinite canvases?

most valuable thing i've internalized as a builder: if you're serious about what you want to ship, stop entertaining every shiny detour that crosses your feed. context switching between frameworks, models, and half-baked ideas is just energy leaking out of a pipe. be unreasonably committed to the problem you actually care about solving.

@cursor_ai very cool achievement, but is grep really the right approach to find information across millions of files? don't such codebases have high duplication of common terms like "cache key", etc?

Cursor can now search millions of files and find results in milliseconds. This dramatically speeds up how fast agents complete tasks. We're sharing how we built Instant Grep, including the algorithms and tradeoffs behind the design.

5.4 is dumb today

software engineering teaches us adaptability, systems thinking, and the discipline to build things that outlast you. it reminds us that even when the tools shift under your feet, the fundamentals of good abstractions and clear thinking still hold.

the hallmark of a good engineer isn't knowing the right answer, it's not letting your current stack become your personality. if you define yourself by your tools, questioning them feels like an attack. if you see yourself as someone who builds things, switching languages or frameworks is just another experiment.

has any human reviewed the questions on swe bench?

@weswinder agreed. for now.

anybody saying you don’t need to look at code anymore is wrong

dear javascript frameworks, please come up with a universal way to say "reject the click if the stuff under the mouse changed less than a second ago"

Wow, I totally disagree with this statement. At the current state, AI actually amplifies the developer to developer difference. If you were a 10x developer, you had good ideas + architectural clarity, this is a brutal advantage when using AI. Steering is a fundamental part of today's AI development.

current status of coding agents: - can do pretty much all minor tasks effectively (up to a day of previous human wall clock) - can do some long horizon stuff unsupervised (week+ of human wall clock). some orchestration glue helps. - still cannot effectively identify and prioritize worthwhile new tasks (but honestly human engineers can't either)

github copilot agents are actually pretty great for small tasks. it's a nice supplementary subscription to claude code or codex (or both)

am I the only one who finds watching agent chain of thoughts endlessly fascinating?

releasing your gstack is like today’s version of putting your dotfiles in a public github repo.

@DevDminGod @garrytan I’d kill to go back to a world in which I couldn’t spot the AI generated tweets down to the specific model used

content and products demand attention of both the consumer and producer. slop has nothing to do with ai, it's when the producer doesn't put in their required share. consumers can tell intuitively when something isn't well thought out, or isn't relevant to them. we've always had a word for this: spam

i've learned this a few times: you can fundamentally rewire how you think about systems and abstraction by spending a month or so seriously building something that is genuinely way outside your current mental model of how software works.

@jeffzwang github.com/gsd-build/get-…

@jeffzwang This is where GSD shines

does anyone have any tips on how to prompt/plan when trying to oneshot large projects, like 50K+ LOC?