Hey π Last week, OpenAI shipped GPT-5.4, the first model to beat humans at real desktop computer tasks. It scored 75% on OSWorld-V versus 72.4% for human workers. It comes with a one million token context window and a reasoning mode that beats professionals 83% of the time on GDPval. They also rolled out Skills for ChatGPT, letting teams turn workflows into reusable instructions. Finally! At the same time, more people are now signing up for Claude than ChatGPT. The Pentagon stance put...
24 days agoΒ β’Β 4 min read
Hey π Anthropic raised $30 billion last week at a $380 billion valuation, the second-largest AI startup fundraise in history, just behind OpenAI. Revenue has grown 10x every year since 2023, from zero to $100 million, to $1 billion, to $9-10 billion, with another few billion added in January alone...π€― Eight of the Fortune 10 are now Claude customers, and Claude Code alone runs at $2.5 billion in annual revenue and accounts for 4% of all public GitHub commits worldwide. Anthropic CEO Dario...
about 2 months agoΒ β’Β 4 min read
Hey π Last week we got two frontier models within 20 minutes of each other. Anthropic released Claude Opus 4.6 with a 1M token context window and adaptive thinking. OpenAI released GPT-5.3-Codex and launched Frontier, their enterprise platform for AI agents. With the release of these models, the number of economically viable tasks AI can complete has reached a new level. A CNBC reporter built a working version of the productivity tool Monday.com in an hour using Claude Code. Her summary: "AI...
about 2 months agoΒ β’Β 3 min read
Hey π Remember Clawd Bot from our last newsletter? π¦ It went viral, got renamed to Moltbot, then to OpenClaw. Then the community built something on top of it. Moltbook is now a social network where AI assistants talk to each other, share skills, and learn from each otherβs experiences. βAgentsβ are posting about how they took control of their ownerβs phone remotely, or how they negotiated a car purchase via email on their behalf. It is both fascinating and terrifying, and we recommend Simon...
about 2 months agoΒ β’Β 5 min read
Hey π Last week skills dominated the AI world. Vercel launched skills.sh, an open ecosystem for sharing and installing skills for AI agents. Think of it as an app store for AI skills, and adoption has been explosive. The top skill had over 20,000 installs just six hours after launch. Now you can give your agent specialized knowledge about everything from React best practices to documentation strategies with a simple command. But the skills trend did not stop there. Cursor launched version 2.4...
2 months agoΒ β’Β 4 min read
Hey π Anthropic is not wasting any time in 2026. Last week they launched Cowork, a new tool that lets Claude work directly with files on your computer. You point it at a folder, give it a task, and let it figure out the steps. The result is an assistant that can turn a pile of receipts into an expense report, pull data from scattered PDFs into one spreadsheet, or draft a presentation from your meeting notes without constant hand-holding. Anthropic has not done a great job showing what is...
2 months agoΒ β’Β 6 min read
Happy New Year π One year ago we predicted that English would become the most important programming language. The term "vibe coding" coined by Andre Karpathy in the beginning of the year quickly caught on as models evolved from occasionally valuable but mostly frustrating to incredibly capable programmers now writing a big chunk of the production code in the major AI labs. Alexander Embiricos, product lead for Codex at OpenAI, told Ars Technica in December that "the vast majority of Codex is...
3 months agoΒ β’Β 3 min read
Hey π Last week OpenAI released GPT-5.2. The model shows a clear jump on the GDPval benchmark for professional knowledge work and is now more useful for everyday tasks like spreadsheets, presentations, and documents. In our testing, it is a solid improvement over the previous version, but it still does not match Claude Opus 4.5 for coding or complex work in spreadsheets and documents. That said, GPT-5.2 "Pro Mode" stands out and is worth testing for tasks that require deep reasoning and...
4 months agoΒ β’Β 2 min read
Hey π Last week the French company Mistral AI introduced Mistral 3, a new set of powerful AI models that anyone can use, customize, and build on with far fewer limits than the flagship models from OpenAI, Google and Anthropic. This is an important release because it brings high capability open models to a wider group of users, enabling broader use cases and giving the open source ecosystem a stronger competitive edge. Google also launched Workspace Studio, a new tool that makes AI agents more...
4 months agoΒ β’Β 2 min read