John Suh
@john_ssuh
Followers
461
Following
155
Media
10
Statuses
84
@ AI Holding Company
San Francisco
Joined October 2025
This should be in SF moma
0
0
1
Or...you can just use TanStack Start
Vercel just dropped 10+ years worth of Claude Code Skills with all their frontend learnings & optimizations 𤯠Ralph is going to love looping this
2
0
2
Claude Cowork is novel for people who haven't seen how people work in the real world, it is strictly less interesting than Manus from last year
0
0
3
https://t.co/Oj2tTgercP Officially skill issue if you think AI is writing bad code. Adjust your mindset instead of dismissing it
Linus is vibe coding. Now that's not what's surprising. Read the last line: "Is this much better than I could do by hand? Sure is." I would invite people to read the entire context before jumping to conclusions, but this is one heck of an endorsement.
0
0
0
https://t.co/0fPW7OnWVm For example Definitely not a junior
The legend himself, Linus Torvalds, just said that vibe coding on the Linux kernel using AI runs better than hand-written code (using googles antigravity) Thatâs a novelty.
0
0
1
A prompt that's been very helpful in reading on https://t.co/fC1SUVZ2av: ```can you transform each paragraph of this chapter to one sentence per paragraph as well? Label the beginning 6 words of the paragraph. Keep specific nouns``` It helps reading the gist of the chapter
0
0
0
I think staff+ get lot of leverage, seniors not much, and juniors get a lot of leverage You can see this based on comments of Redis founder @antirez Planetscale CEO (ex Vitess) @samlambert Railway CEO @JustJake Linus @Linus__Torvalds George Hotz (post CC) @realGeorgeHotz
"juniors like theo report huge productivity gains from their low starting point" @theo grok fucked you on this one
1
0
2
https://t.co/YvC76rngT2 The word of the goat
anthropic rugging claude code access has made me realize i prefer codex with Clawd bot.
0
0
16
@Bootlegregore I was the biggest fan of cc when it first came out, but I feel that there is perhaps an internal tug of war between trying to optimize for power user delight who wants to synchronously drive the agent vs correctness in long complex task in yolo mode I couldnât care less for the
0
1
6
Because people are pretty slow on Twitter, it might take 6 months for people to realize that 1. Codex is better than Claude Code (gpt-5.2 high v Opus) 2. OAI models are much smarter than Claude models at doing any type of work with thinking, but Anthropic models have better
With the Claude Code shutdown. I am proud that we build Codex in the open with our OSS repo and we are 100% invested in supporting a flourishing ecosystem of agentic coding tools out there. You can already build on top of https://t.co/AwDBgSWVqU directly, which includes ChatGPT
165
92
1K
Quick demo of Alexandria0âs core feature: âI instantly loads highlighted EPUB text into chat so you can ask questions in context. New: - âK for fast navigation - Right-click â âAdd to chatâ
1
0
8
Try at https://t.co/fC1SUVZA03 Currently loaded with 10,000 Gutenberg books. $0 - you do need to bring your own Gemini API Key to chat.
0
0
1
Before AI, it would've been simply impossible to ship this hobby project because there are so many features to support which you can explore: 1. An actually nice-looking, functional web-based epub reader (inspired by Books app on MacOS) with global theming, left / right arrow +
1
0
3
I made a "Cursor" for epub in 4 days, 100% coded with AI (@OpenAI Codex 90%, @AnthropicAI Claude Code 10%)
3
3
32
Claude Skill marks the end of OpenAI Agents, Zaipier, Langflow, Retool Workflow, and all other similar products - you simply need to define the integration and a reusable prompt instead of building clunky workflows - refer to @bcherny recent replies
1
0
0
âBackground agent infraâ will be the topic of 2026
yes things are changing fast, but also I see companies (even faang) way behind the frontier for no reason. you are guaranteed to lose if you fall behind. the no unforced-errors ai leader playbook: For your team: - use coding agents. give all engineers their pick of harnesses,
9
9
206
It seems like when @karpathy did the interview with @dwarkesh_sp, he had only used Cursor and wasnât fully aware of the state of the art of agents. I predict a new podcast in the next 6 months where he drastically shortens his timelines and expands the impact of agents as-is
2
1
4