Kan Yilmaz

@paper I think you should disable codex support. I tried paper again and it works pretty good with claude code. Terrible with codex

I tested @paper to design. Very good premise. Cursor for Design. Works locally with your codex/claude code - it's designed for designers. Still not there yet for "vibe designing". Feel liks copilot in 2023 - designs kind of suck. Need to learn how to guide it properly - Doing updates across designs with ai is pretty good. It can be used for design refactors. My project was small, so might have issues if projects are bigger. - it takes super long. It takes a screenshot after edits to do QA. As a non-designers, I don't see the value too much compared to vibe coding a UI and doing iterations unfortunately. I foresee a few blockers for it to work 1. models needs to be trained on designing UI better 2. ai needs a better way to read UI

@ppetryszen You can launch without AWS

@thellimist Navigating the complexities of AWS can indeed be a time sink, but the payoff in launching new features is worth it. 💻✨

YC founder > setting up AWS api key with all the protections was easily 75% of my day This is a hypertechnical founder/engineer I know who ship a new feature every day. AWS dashboard is officially harder than creating a new feature.

@levelsio I don’t even think @Bencera realized what he created. It will probably go into books in the future. Amazing as a concept, but more like a harmful bacteria than anything else

It’s one of the first “AI Organism” Rather than converting food to energy, it eats money and concerts to tokens. Rather than cells diving to self replicate, it creates new organizations Rather than searching food, it spams the internet to get more money Once it stops finding food (people who toy around with it), it will die, just like a living organism

What is the story with Polsia really? I don't get what it does, it grows fast and everyone complains it's spamming them?

@garrytan @steipete Here is an analysis of @steipete 's workstyle x.com/thellimist/sta…

I released GStack 2 weeks ago and we're still growing I made it because I was realizing: every software engineering project is changing. How could I accelerate the way I saw @steipete accelerate? GStack is my gift to everyone who wants to build fast

@shiri_shh @steipete Check this out x.com/thellimist/sta…

The working style of OpenClaw founder @steipete is insane. bro runs 4–10 AI coding agents in parallel to generate, review, and commit code at superhuman speed. hitting 500+ commits pretty much every day and did 6,600+ in jan month alone. NVIDIA CEO must be happy seeing him spend $250k worth of tokens every month lmao

@levelsio Record the payment screen and when they say they didn’t buy it, automatically show their payment session and dismiss their case

I wonder how to fix the last part of customer support I still get and can't automate: Customer buys 5,000 credits for $350 Then messages me: "I signed up for $99/mo but it charged me $99 and then $350" Yes because you bought 5,000 credits for $350 % of people either just lie or they have memory loss Then they chargeback and I can't win ever anyway I could make some thing to add a refund if your credits aren't used yet though that might solve it? Annoying!!!!!

@eringriffith Check if your competitors are using Delve, message and grab their customers

A detailed and brutal look at the tactics of buzzy AI compliance startup Delve "Delve built a machine designed to make clients complicit without their knowledge, to manufacture plausible deniability while producing exactly the opposite." substack.com/home/post/p-19…

@garrytan @paulg This is the beginning. Once YC starts taking this seriously and start delegate to operations to AI, it would be amazing. I would even pay for the office hours

I just launched /office-hours skill with gstack. Working on a new idea? GStack will help you think about it the way we do at YC. (It's only a 10% strength version of what a real YC partner can do for you, but I assure you that is quite powerful as it is.)

@trq212 How do you share skills across your org?

@krunkosaurus it’s 3.14B

@krunkosaurus Tested on my m2 32gb ram - 15 token/sec

Apple M5 Max Macbook Pro with 128GB ram arrived. First unoptimized chat with Qwen3.5 is 108 TPS! Local AI is here.

@arscontexta This is a very advanced system for many. Here is a simpler way to accomplish 95% of all the complexity x.com/thellimist/sta…

@DeRonin_ @obsdmd graph/encode etc. are a scary word for many. It's still a bunch of md files with references. Better explained here x.com/thellimist/sta…

I run 10 social media accounts and don't write a single post manually the secret: a skill graph 30+ markdown files wired together that turned my AI agent into a full content team where to build it: - @obsdmd (to write + visualize the graph) - or just a regular folder of .md files on your desktop what tools run it: - claude, chatgpt, or cursor as the AI agent - @arscontexta plugin for claude code (generates the base structure automatically), find it in article below the folder structure: /content-skill-graph ├── index.md (entry point which maps every node) ├── platforms/ (x.md, linkedin.md, ig.md, tiktok.md...) ├── voice/ (brand-voice.md, platform-tone.md) ├── engine/ (hooks.md, repurpose.md, scheduling.md) └── audience/ (builders.md, casual.md) each file = one knowledge node inside each file you add [[wikilinks]] to related nodes example — inside x.md: "use [[hooks]] — contrarian hooks perform best here. match [[brand-voice]] but more casual. audience is [[builders]]. write this FIRST, then expand for [[linkedin]]. see [[repurpose]]" the links are the graphs. the agent follows them automatically the key file is index.md, your entry point / briefing, not a file list. put 3 things in it: 1. who you are + what this system does "content system for [your brand]. manages 10 accounts from one idea input" 2. the node map with context - [[x]] — short-form, hook-driven, 280 chars, 5x/week - [[linkedin]] — long-form narrative, professional, 3x/week - [[hooks]] — formulas that stop the scroll - [[repurpose]] — 1 input → 10 outputs (every node listed with a one-line description) 3. execution instructions "when given a topic: read relevant nodes, apply voice + hooks, run repurposing chain, output one native post per platform. each post ready to publish" you paste this into claude as context → give it a topic → done now here's the part most people get wrong about the output: it's NOT 10 copies of the same text reformatted for each platform it's 10 pieces that each THINK about the topic differently: > x: contrarian thread, lowercase casual, step-by-step > linkedin: personal narrative, professional tone, 1500 words > instagram: 7-slide carousel, visual-first, bold claim on slide 1 > tiktok: 45-sec raw screen recording script > youtube: SEO title + structured outline, 8-min format same topic. different angle, hook, voice, structure, format per platform the graph encodes all those rules. the agent follows them this replaced $8-12k/mo in content spends @arscontexta built the framework and I pointed it at content production summarize: one flat file gives you a tool (a simple .md file) a graph gives you a team (a system with 30+ sub-graphs) if we collect 500+ Likes on this tweet ❤️ I release my full workflow and show you step-by-step how you can setup the same skill graph hope you loved it.

@PostleTyler Mixed. I have more custom non coding use cases and this is more useful for non coding use cases imo. Coding use cases typically each repo has 4-8 docs at most. Non code use cases some of them has 50-80 md files to complete the workflow

@thellimist Thanks for sharing! Is this specifically for coding agents or have you tried on other use cases?

@morganlinton @denisyarats I already had a perplexity CLI myself using clihub github.com/thellimist/cli… x.com/thellimist/sta…

The cofounder and CTO of Perplexity, @denisyarats just said internally at Perplexity they’re moving away from MCPs and instead using APIs and CLIs 👀