Anthropic makes its first acquisition
Make your own benchmarks
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,
I’m launching a fun little thing: droidmas.com - jokes and games around concepts in ai dev. Day 1 is about benchmarks, live now.
Anthropic has acquired Bun. Bun makes JS code run faster, and the company is filled with the best JS devs. The story of how this came to be is crazy. OpenAI also bought a company, Neptune AI (tools for AI researchers).
Opus 4.5 is finally available on Claude Code with Pro plans ($20/mo). The limits are not as generous, but at least you can use it :) Claude Code has crossed $1B in run-rate revenue just 6 months after launch, and the rumours are that Anthropic is planning an IPO.
Google Workspace Studio is now GA. It’s an automation builder, but it looks like it can handle much more complex use cases (connect multiple apps beyond Google’s ecosystem, add different AI steps in between), all with just prompts.
Mistral released a few models: Mistral Large 3 and Ministral 3. Mistral Large 3 is in the same leagues as DeepSeek 3.2 and Kimi K2. I’m more excited by Ministral models. They come in three sizes: 14B, 8B, and 3B with base, instruct and reasoning variants for all three.
Eliminate AI anxiety. Build with confidence. AI agents are the future—but centralized systems create costly risks. With Airia, prototype and orchestrate AI agents securely while creating a resilient, decentralized AI ecosystem for your organization. Get started today at airia.com*
🌐 What I’m consuming
How is AI changing work inside Anthropic? Basically, engineers feel ~50% more productive, 27% of Claude-assisted tasks were projects that “would not have been done otherwise.
Lessons from building AI agents for messy government PDFs.
What I learned building an opinionated and minimal coding agent.
One year of using ChatGPT Pro as my first hire.
Cursor’s guide for non-technical hires to build with Cursor.
⚙️ Tools and demos
Bloom - Distil your website into a living brand system and create on-brand assets.
Exa handles simple and complex queries from the same search box.
Shortcut - Superhuman Excel Agent that takes you from beginner to pro.
Braintrust (evals platform) added a feature that lets users vibe code custom annotation UIs directly on the platform.
Anything Max - A $200/mo plan that comes with autonomous backend and frontend testing to actually ship your vibe coding projects.
Silvia – Turns your entire balance sheet into a live AI dashboard. Tracks accounts, models “what if” scenarios, and surfaces insights automatically.*
🍦 Afters
New research from OpenAI trains a version of GPT-5 to confess when it has made a mistake.
In red teaming research from Anthropic, AI models found $4.6M worth of exploits in blockchain smart contracts.
Building coding agents with tool execution - new short course from E2B and DeepLearning AI.
Apple’s head of UI design is leaving for Meta to oversee AI and hardware integration.
That’s it for today. Feel free to comment and share your thoughts. 👋
Read about me and Ben’s Bites
📷 thumbnail creds: @keshavatearth,
Thanks to today’s sponsors who made this newsletter possible :)
Wanna partner with us? Last few slots left for the rest of the year.