Kyle Stewart-Frantz

@charliermarsh with pydantic support 😅

Next stop ty stable release

meta large scale plan - dispatch subagents for read and write - make sure plan is in phases. do a phase, update plan - deprecate old phase and continue on. did 12 hour session yesterday with 60 some odd sub agents and one shotted. when code doesn’t muck up the context window much easier.

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

@theo no sonic in your area?

Shit.

@guidorosso @rive_app nice on the role!! about time was cheeky to growing in the direction. much love !

@kyle_frantz @rive_app See thread for a lot more examples

Hiring alert: Applied AI Engineer role at Rive, please help share/repost! apply.workable.com/rivecareers/j/…

@stoolpresidente gotta hit up casey pizza in mission bay

Barstool Pizza Review - Flour + Water Pizzeria (San Francisco, CA)

@exec_sum so one martini?

BREAKING: Goldman Sachs raises dinner allowances for employees working past 8pm from $30 to $35.

@stoolpresidente try casey’s in mission bay

Barstool Pizza Review - Square Pie Guys (San Francisco, CA)

@FrankieIsLost banger

i miss the old opus one-shot the code opus call the right tools opus go straight to prod opus

@karpathy x.com/kyle_frantz/st…

Finding myself going back to RSS/Atom feeds a lot more recently. There's a lot more higher quality longform and a lot less slop intended to provoke. Any product that happens to look a bit different today but that has fundamentally the same incentive structures will eventually converge to the same black hole at the center of gravity well. We should bring back RSS - it's open, pervasive, hackable. Download a client, e.g. NetNewsWire (or vibe code one) Cold start: example of getting off the ground, here is a list of 92 RSS feeds of blogs that were most popular on HN in 2025: gist.github.com/emschwartz/e6d… Works great and you will lose a lot fewer brain cells. I don't know, something has to change.

@samuelcolvin @claudeai You could also consider running hourly time machine or syncthing so that there is file redundancy

My first really serious fuck up of @claudeai code. Tried to get it to fix my ghostty setup. It nuked my .zshrc I've been carefully curating for years. Just rewrote it to contain two lines it thought might fix the problem. I happen to have a random backup from last June, no idea what's in it.

article fatigue 🙁

@TheAhmadOsman 12x3090s vs 1 RTX Pro 6000 if you had to buy at once.

looking to Buy a GPU and run your LLMs locally? I got you read through this thread and you’ll be able to build an AI server up to 16x GPUs & run the latest State of The Art LLMs at home * Will be answering all ur hardware questions throughout the weekend on this tweet, ask away

@RhysSullivan 💀

opus 4.5 is incredible

@mweinbach pretty sure i saw one in wyoming last week

This is Nvidia’s end to end autonomous vehicle model running in a Mercedes car going point to point through San Francisco with no interventions, one shot

i miss the old days when we craft code by hand. hits different now.

@petergyang @cursor_ai github.com/jonathanyeung/…

.@cursor_ai should just build the ability to view formatted md files and edit them directly (or maybe this exists?). I don't like using another separate app (Obsidian)

Is anyone using Claude Code or similar for stuff that isn't coding at all, purely because it's nice to have access to local files and stuff? I use AI as a thinking partner constantly, and not having persistent references is starting to become annoying, but worried the CLI tools are tuned too much for programming and not business strategy for example.

@TheAhmadOsman @TheAhmadOsman you have to try the gorilla mind drinks

if you don’t pick up a 24 pack of energy drinks for the holiday season and prefer to spend time with your family, you’re not ngmi