@_brimtown @rough__sea thanks!!
i definitely agree that legacy code is only tech debt if you have to maintain it. if it's a throwaway prototype or simple enough that you can literally re-generate it from scratch for every change, then yes, vibe coding is awesome. i do it all the time and recommend it!…
Notes
Text may be partial · Read original
- @brandonwilson @rough__sea I'm more a founder than an engineer at this point, so I think I take the entrepreneur's point of view What's a good concrete example of how AI is fundamentally changing how engineering is done? (ie, more than just punchcards to textual programming, which to be fair is a big…
Text may be partial · Read original
- @chieflybrit "always bet on text" text is so easy to ingest and output and there's so much of it, so high signal, low data (ie high bang for buck density) the bitter lesson, all the code on github, the internet to train on
- immaculate vibes across the board 👏 @posthog https://x.com/stevekrouse/status/2013295171832573964/photo/1
- @brandonwilson @rough__sea I haven't seen that yet, but maybe! Have you gotten a chance to watch my talk?
- @rough__sea This is true in the same way that the era of punchcards came to an end. The way we write code is definitely changing! The real question is what will be the role of the human brain in programming? I gave a talk about it 6 months ago that I think holds up: https://www.youtube.com/watch?v=1WC8dxMC4Xw
- In retrospect, it seems pretty obvious that the first AI systems would produce text and get better at software faster than driving cars It also seems obvious that we'd want it to produce JSON to hit APIs Agent loops with tools also seem obvious I feel like Asimov or Minsky could've predicted all this in the 1940s Yet it feels like we all figured that out together over the last 6 years I wonder what else will be similarly obvious in retrospect Maybe with some clever thinking we could think out a few steps ahead?
- RT @jmduke: 1. There are no big secrets, just an infinite amount of tiny ones. 2. Being 20 is a superpower; don't take energy and naiveté…
Reposted by Steve
Text may be partial · Read original
- I quoted @phokarlsson twice!
- https://stevekrouse.com/advice
- I have some college students visiting my office this week, so I put together a list of "advice to college students"! It was fun! https://x.com/stevekrouse/status/2012982999311069413/photo/1
- I'm going to be meeting with a bunch of college kids, and I'd love to give em some good wisdom What do you wish someone told you when you were in college? What would you tell college students in 2026?
- I'm going to be in SF the last week of Feb Any interest in a F̶u̶t̶u̶r̶e̶ ̶o̶f̶ ̶C̶o̶d̶i̶n̶g̶Feeling of Computing meetup? Any suggestions on location?
- RT @kamalnrf: I do this with @ValDotTown + Claude. The way Vals need no setup or deploy step feels instantaneous. My prompt is usually: “c…
Reposted by Steve
Text may be partial · Read original
- Without looking it up, how many miles is the earth from the sun?
- @geoffreylitt @nileshtrivedi @worrydream @michael_nielsen same!
- @artivilla I'm confused. The MCP is an interface. Could you ask a different way?
- An MCP is an API you can change without breaking backwards compatibility. An LLM can figure out how to call it afresh every time So we shove more features into our MCP. Which makes it a tempting surface to program against. Which means folks will be upset when we break it 😅 Anyways, thanks @kamalnrf for making a Val Town MCP SDK so you can call our MCP as an API https://www.val.town/x/kamalnrf/valtown-mcp-client Who knows? Maybe one day soon we'll all sprinkle a little runtime LLM inference into all our API calls
- @thenanyu ha! i was 80% sure i'd see you here ;)
- Has anyone had a good experience with a "github issues agent" that brings in good relevant context from existing issues or other codebase knowledge? I imagine a good product in this space would need a fairly sophisticated setup to efficiently store all the knowledge in a repo, including the commit & pr history, but maybe there's a simple architecture that works ok?
- @NicolaeRusan PDFs? https://x.com/stevekrouse/status/2002466269141086311?s=20
- @samlambert You personal spend 25% of your time sourcing??? That's awesome!
- RT @geoffreylitt: “Human review as the bottleneck for AI work” suggests an *education* problem I don’t just want a code diff and some test…
Reposted by Steve
Text may be partial · Read original
- @Prigoose do you have one?? does it work? any downsides?
- @danshipper how to summon the ghost of christmas past
- @MaineFrameworks @worrydream It's where logo came from
- "Mindstorms" by Seymour Papert is the best book ever written about education & learning. An MIT scientist created a radioactive turtle to bite children and give them mental superpowers. I was bitten when I was 12. Before the bite, I hated math, and was mediocre at school.… https://x.com/stevekrouse/status/2003063621631271345/photo/1
- Has anyone turned Mindstorms into an audiobook? Asking for a me
- My pagerduty sound is a cat meowing So now when my servers are struggling, my fiance dreams of cats
- Do you know what a UAP is?
- @Willthereader @nbbaier Emily
- @Willthereader @nbbaier Yes
- @nfarina fun idea!
- townie v5 has been achieved internally what should i ask it to do?
- @abstractalgo Yes, you are trying to go North But you don't have a good map So you need to scout out parts of the terrain to map it But each part of the map you need to scout is itself full of "sub-maps" which admit the same problems as the top-level map
- Prioritization feels like it should be choosing where to go on a map But you only have a partially accurate map So first you have to decide which shrouded spots on the map you want to explore and map out Then once you've mapped out some spots, then you can decide where to go,… https://x.com/stevekrouse/status/2000602889530585530
- Project prioritization seems like it should just be ranking projects by impact vs effort But the list itself is fluid: exploring a problem by coming up with clever solutions can change the impact vs effort But exploration itself is costly too. You can't afford to explore every problem indefinitely So the real challenge of product management is prioritizing where to spend your exploration budget, so you can then you can prioritize which projects to work on
- @nbbaier I always have students. I'm a teacher at heart
- "What is a 'console'? Why would I want to 'log' to it? What is a 'log'?" Very good questions from my newest student
- RT @Paul_Kinlan: It was fun to build https://flickity.val.run/ I turn each article linked from hackernews into a short video by summarizing…
Reposted by Steve
Text may be partial · Read original


