Raghav

Immersed in AI building apps

[Boris Cherny @bcherny]

In the next version of Claude Code: run /usage to see a breakdown of which Skills, Agents, MCPs, and Plugins are using your tokens

CLI today, coming to Desktop next

[Hasan Toor @hasantoxr]

This is the biggest LLM breakthrough in years.

SubQ IS the first frontier model with a fully sub-quadratic sparse-attention architecture (SSA) + a 12 MILLION token context window.

The numbers are insane:
→ 52x faster than FlashAttention at 1M tokens
→ Less than 5% the cost of Opus
→ Nearly 1,000x less compute by ignoring the 99% of token relationships that don’t matter

No more quadratic waste. No more context hacks.

Early access is open right now (plus their new coding agent SubQ Code).

This is how LLMs actually scale next.

Get it → https://t.co/FKHX1vPRhI

[老鬼 @laogui]

GitHub 联合创始人 Scott Chacon 的新项目 GitButler 刚刚宣布完成 1700 万美元 A 轮融资！

GitButler 是一款为现代 AI 编程工作流打造的创新型 Git 客户端（支持桌面端、CLI 以及全新的终端 TUI）,专为 AI Agent 时代设计 — 不只是给人用，也给 AI 用。

它不是"更好的 Git"，而是在重新思考下一代软件应该怎么被构建。

它的核心优势在于打破了传统的分支工作流，支持并行分支（Parallel Branches）和堆叠分支（Stacked Branches）。开发者可以在同一个工作区同时推进多个功能的开发（例如边修 Bug 边做新功能），彻底告别繁琐的 stash 和分支切换。此外，它深度集成了 Claude 等 AI 助手，能自动生成分支名、提交信息，并支持无限撤销与轻松的提交编辑。

在 UI 层面，GitButler 的设计极具现代感与实用性：采用独特的水平滚动视图，将未分配的更改、多个虚拟分支并排清晰展示，视觉层级分明。

[Graeme @gkisokay]

Anthropic just banned Claude subscriptions from powering OpenClaw.

Here's why my stack was already built for this.

I never ran Opus 4.6 through a subscription for OpenClaw or Hermes. It runs in Claude Code for complex external dev only. Same with GPT-5.4 in Codex.

The internal agent runtime is a completely different stack:

1. Qwen3.5 9B runs locally. $0. Always on. Feeds the subconscious ideation loop 24/7. Beats GPT-OSS-120B by 13x. Awesome.

2. MiniMax M2.7 is the agent's backbone. 97% skill adherence, built for agents, $0.30/M tokens. The $10 plan allows for 1500 calls every 5 hours. Amazing.

3. GPT-5.4 mini is the Hermes brain. debates ideas with the subconscious, builds output, ~$0.075 avg per run. It's smart enough to orchestrate your entire system, and you can actually use your subscription plan here via OAuth. Incredible!

Over the last 24 hours, the subconscious ran 15 times, for a total of $1.58. Not too shabby for an always-improving agentic system.

The lesson is to build your agent stack on a multiple LLM stack.

Local models handle volume. Generous subscription models handle execution and judgment. You own the cost structure.

Full-stack breakdown in the table. (see image)

[Gauri Gupta @gauri__gupta]

https://t.co/gpEE2rstjm

[PrismML @PrismML]

Today, we are emerging from stealth and launching PrismML, an AI lab with Caltech origins that is centered on building the most concentrated form of intelligence.

At PrismML, we believe that the next major leaps in AI will be driven by order-of-magnitude improvements in intelligence density, not just sheer parameter count.

Our first proof point is the 1-bit Bonsai 8B, a 1-bit weight model that fits into 1.15 GBs of memory and delivers over 10x the intelligence density of its full-precision counterparts. It is 14x smaller, 8x faster, and 5x more energy efficient on edge hardware while remaining competitive with other models in its parameter-class.
We are open-sourcing the model under Apache 2.0 license, along with Bonsai 4B and 1.7B models.

When advanced models become small, fast, and efficient enough to run locally, the design space for AI changes immediately. We believe in a future of on-device agents, real-time robotics, offline intelligence and entirely new products that were previously impossible.

We are excited to share our vision with you and keep working in the future to push the frontier of intelligence to the edge.

[BuBBliK @k1rallik]

https://t.co/uAaxXKPqnk

[alex @kvamme]

Stripe Projects launching with @trychroma. The benefit of being the OSS leader is that you start becoming the default in the agent ecosystem. Excited for this.

[Philipp Schmid @_philschmid]

https://t.co/diF2swHIVb

[Sam Rose @samwhoo]

I think this is the best post I've ever made.

[Ben Cera @Bencera]

$6M run rate. $3M->$6M in 2 weeks. One Founder + AI agents. Zero employees.

I wanted to create a platform with the vibes of the 1990s, the vibes of the 2000s, of the 2010s, and then have a feature of the future

And I said, "Wait a second, I know the Agent SDK
Why don't I use the Agent SDK which is the feature of the future?"

And I didn't have any idea what to do, but I knew I needed agents, so I put agents in loops and connected MCPs, which then were synced to real products running in production

I knew that could be a feature of the future but I didn't realize how much the impact would be

[Raghav @engfounder]

Every employee is an AI agent. Every job description is a Markdown file. There's no framework, no SDK — just .md files and Claude Code. @garrytan
I broke down the entire architecture:
https://t.co/Y63HHGYscZ

[Raghav @engfounder]

https://t.co/nf2Mu8qZuW

[Raghav @engfounder]

https://t.co/nf2Mu8qZuW

[Chrys Bader @chrysb]

introducing AlphaClaw Apex 🐺

a native Mac app for managing multiple OpenClaw VPS instances from one dashboard.

some of you are already setting up OpenClaw for clients as a service. Apex is built for you.

deploy to Hetzner VPS in one click. monitor all your instances. manage configs, updates, spend, and health from a single UI. no SSH needed.

everything you know from AlphaClaw, now across a fleet:

📅 Google Workspace OAuth & pubsub wizard
⏱️ Cron calendar view and cost-saving insights
🖥️ Remote node setup wizard
🔄 Auto-backup to GitHub
📊 Token usage & cost analytics built in
🧱 Prompt hardening reduces agent drift
🩺 Drift Doctor analyzes your prompts and workspace for drift
💬 Telegram multi-topic workspace setup wizard
📂 Full file browser, editor, and terminal no SSH needed
🐕 Per-instance watchdog and crash recovery
🛠️ Manage env vars from the UI
🔑 Manage model keys & OAuth visually
🪝 Webhook creator & inspector with replay & debug
⬆️ One-click updates, no redeploy needed
📦 Import existing setup from GitHub

if you're thinking about offering managed OpenClaw as a service, this is the ops layer you've been missing.

[Garry Tan @garrytan]

It's working guys. This is exactly what I built for myself and what you can have for you now.

It's also free and open source and you can fork it and make it your own. I don't want to hear the hate. This is a gift!

[Raghav @engfounder]

@claudeai seems to have problems with Claude Desktop app. Timing out and no chat is getting submitted.

[Garry Tan @garrytan]

I am coding a lot, GStack is helping me do it, but also I want you to know I was stranded in Austin the last 24 hours due to weather, and also last week my mom was in the hospital and not too lucid for most of it, so I was coding by her bedside too. She's ok now and I just visited her at home and set up her medication.

I do have a full time busy job, and is it really possible for a CEO to be coding all the time? Frankly, I think it will have to be. The CEO has to set the future of the company. All companies will need to adapt to a faster world and do more. Boil the ocean. It's not about doing less and cheaper. It's about doing more and making 10x better products and services.

Is 16k LOC/day sustainable for me? We're going to find out if I can manage to get to L8 software factory. I have not done it yet.

But you can tell the models are about to get much much better. L8 is barely possible today, and I think I'm close. But everyone will be there soon.

I want to be one of the people who helps all of you do it with me.

[Garry Tan @garrytan]

I've been having such an amazing time with Claude Code I wanted you to be able to have my *exact* skill setup:

Introducing gstack, which you can install just by pasting a short piece of text into your Claude code

[akira @realmcore_]

https://t.co/CudCkyWuRs

[weisser @julianweisser]

$1.5M ARR in 2 weeks.

Zero (human) teammates.

Solo Founders Podcast is live with @bencera of @polsia, an AI that runs your company while you sleep.

His solo founder rule: 80% AI, 20% taste.

0:00 — How "solo founder" has changed with AI
2:40 — Scoping and trusting AI agents to ship
5:30 — Cloud Kitchens with Travis Kalanick
9:52 — Mount Fuji: Where Polsia was born
13:55 — 80/20 rule: 80% AI, 20% taste
21:24 — Building for yourself, not imaginary customers
23:10 — The game that inspired Polsia
29:55 — Polsia as an economy: The bigger vision
40:59 ��� How Polsia actually works today
51:35 — Why not just build businesses yourself?
57:55 — Advice for new builders: Push AI to the edge