The newsletter for the technically curious. Updates, tool reviews, and lay of the land from an exited founder turned investor and forever tinkerer.

Hey folks,

While we’re all waiting for Gemini 3 (probably out when some of you are reading this 👀) , Grok 4.1⁶⁷ is out. It’s on top of LM Arena¹⁵³, has good creative writing and comes in two variants - thinking and non-thinking. There are no other “performance” benchmarks, but the model safety card mentions that Grok 4.1 has much higher levels of sycophancy (i.e. agreeing with the user) and deception than Grok 4. I tried some basic prompts, and it looks like a decent model, but I don’t feel like I’m getting something special to even consider using it regularly. (Like every Grok model tbh)

OpenAI made GPT-5.1 available in the API⁴⁸ and shared some evals for it. On vibes, I’m seeing reports of the model being “extra eager”, similar to Claude Sonnet 3.7.

OpenAI is also piloting a group chat feature in ChatGPT¹⁷¹ in 4 countries: Japan, New Zealand, South Korea and Taiwan. Poe⁵⁸, the AI chatbot from Quora, released the same feature with groups up to 200 members, including humans and multiple models.

Anthropic says it stopped a large-scale AI cyberattack⁸⁹ (originating from China), open-sourced their eval for political bias⁷⁶, and launched a learning companion⁶⁹ with the Govt. of Rwanda. Oh, and the Claude Code team released a pre-built plugin to make it better at frontend design.²³¹

Attio¹⁸⁶ is the AI-native CRM for the next generation of teams. Sync your email and calendar, and Attio instantly builds your CRM—enriching every company, contact, and interaction with actionable insights in seconds. Join fast growing teams like Granola, Flatfile, Modal, and more. Start for free today¹⁸⁶.*

Inside Factory

• Our CLI now supports all Skills¹³⁰ - you can import from claude code, plugin support coming this week 😊

• We support Hooks⁶⁴ (which I’ve been experimenting²⁸ with) too

• We have 5.1, 5.1 Codex, and some of the newest models (with SOTA scores) coming out VERY soon

• We’re building a new web product we’re excited to share soon for early access

• We launched a $2k/mo plan, which isn’t as nuts as you’d expect - its actually a KILLER deal for small teams, and I can tell you that people have already signed up to it (one guys’s a solo dev!)

• Want to be more technical? Come join the Factory Discord⁵¹

🌐 What I’m consuming

• The rise of Gamma⁵⁴¹ from “dumbest idea I’ve heard” to $100M ARR. (Luckily, I didn’t think it was a dumb idea and wrote a check in 2022 as a a16z scout - who led their last round)

• Why AI writing is mid⁵⁵³.

• How we made sandboxed coding agents 10x faster to start¹³⁰.

• The state of Chinese LLMs²⁰⁴ - supposedly frontier performance, supposedly huge shadow adoption, all under massive compute constraints.

• How three YC startups⁵¹² built their companies with Claude Code.

• Benchmarks should focus on agentic work⁶⁵ and brittleness⁸⁷, and why this new one isn’t a good measure for hallucination⁴⁴.

⚙️ Tools and demos

• Voice agents fail in noisy, multi-speaker environments. Speechmatics voice API doesn’t. Build with $200 free credits 👈⁶⁶*

• ElevenLabs Image & Video¹⁵⁶ - Generate visual content with the best models (like Veo, Sora and more, then add voices and sound effects in one place.

• NotebookLM¹⁶⁰ also has a Deep Research mode now, with support to bring your own sources.

• React Grab¹⁵¹ - Select elements and edit with Cursor/Claude Code.

• Researchoor¹⁴² - AI-powered social listicle builder.

• Typeless¹⁹⁹ - Convert spoken voice into polished writing.

🥣 Dev dish

• git-worktree-runner⁹⁴ by CodeRabbit - A CLI tool for managing git worktrees with ease.

• ai-sdk-tools/ocr⁹⁶ - Extract structured data from invoices & receipts with one line of code.

• browser-tools¹³¹ - A tiny script that can do most of what chrome-devtools MCP can.

• Oracle⁷⁷ - bundle a prompt plus the right files and hand them to GPT-5 Pro when you’re stuck.

🧙 AI experts you can hire

ICYMI, we built a platform to hire the best AI experts²⁰⁴ for your projects. Here are a few of them:

• Chris Hennigfeld¹²⁷

• Tommy Wu⁹¹

• Alexandra Zubko⁷⁵

• María Kolodynski⁶⁰

• Nick Winder⁶⁶

• Jens jk⁵²

• Alexander Orlov⁶⁵

If you’re a company needing AI projects built, you can post jobs and connect with the right people. We’ll be trialling a personalised matching service soon too.

🍦 Afters

• Cloudflare has acquired Replicate⁶², which gives developers an easy API to build on top of open-weights models (primarily media generation).

• SIMA 2 from Google⁷⁷ - Agent that plays, reasons, and learns with you in virtual 3D worlds.

• Jeff Bezos is starting a new “AI for manufacturing¹⁷⁹” company with $6.2B of funding and himself as co-CEO.

• OpenAI is allowing employees to donate equity to charity⁴³.

That’s it for today. Feel free to comment and share your thoughts. 👋

• Find me on X¹⁶, Linkedin³¹, or Instagram¹⁸

• Read about me²⁴ and Ben’s Bites

• 📷 thumbnail creds: @keshavatearth¹⁹,

Thanks to today’s sponsors who made this newsletter possible :)

Attio and Speechmatics.

Wanna partner with us¹⁴? Last few slots left for the rest of the year.