AI Mornings with Andreas Vig
Your daily AI news briefing in under 10 minutes. New models, product launches, research breakthroughs, and industry shifts, explained clearly, no hype.
AI Mornings with Andreas Vig
Claude's Office Takeover & AI Solves 42-Year Math Problem
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Hey, welcome to AI Mornings with Andreas Vig. It's April 14th, 2026. Claude is continuing its push into the enterprise. Anthropic just announced Microsoft Word integration, letting Claude read comments, make edits, and automate workflows directly inside your documents. The feature also lets you save common workflows as reusable skills. It's in beta for team and enterprise plans, following earlier integrations with Excel and PowerPoint. This is part of a broader pattern. Anthropic is building deep hooks into the Microsoft Office ecosystem, making Claude increasingly sticky for enterprise users. Speaking of the AI arms race, AMD just entered the agent framework game. They've released Gaia, an open source framework for building AI agents that run entirely on local hardware. No cloud dependency, no API keys, no data leaving your device. It supports both Python and C SDKs, and it's optimized for AMD's Ryzen AI chips. The framework handles document QA, speech-to-speech interaction, code generation, and connects to external tools via MCP. This is a strategic play for privacy-first and air-gapped environments where cloud AI simply isn't an option. In acquisitions, OpenAI just picked up Hero Finance, a personal finance startup founded by Ethan Bloch. Hero offered AI-powered financial planning that modeled different scenarios based on your salary, debts, and expenses. The company was specifically trained for financial math accuracy. Bloch previously founded Digit, which he sold for about 230 million US dollars. This looks like an Aquahire Hero is shutting down operations on April 20th with the team joining OpenAI. Interesting timing given Perplexity's recent push into personal finance. In research news, there's a fascinating feature out of Quantum Magazine about AI's role in mathematics. After AI models solved five of six problems at last year's International Mathematical Olympiad, mathematicians started taking them seriously for research work. The article documents several breakthroughs, including UCLA's Ernest Ryu using ChatGPT to prove a 4-2 year old open problem in optimization theory that had stumped mathematicians since 1983. Terence Tao says AI is enabling mathematicians to solve thousands of problems at once and do statistical studies that weren't possible before. He describes the current moment as a lot of throwing things at the wall to see what sticks. Several mathematicians are now leaving academia for AI-focused startups. There's also a new benchmark worth knowing about. 4 currently leads at about 84% with GLM 5. One and Claude Opus 4.6 close behind at around 80%. All traces are publicly browsable, which makes this particularly useful for understanding real-world cybersecurity capabilities. On the infrastructure side, Vercel's CEO Guillermo Rauch is signaling IPO readiness. The company's annual recurring revenue has grown from 100 million to 340 million US dollars in just two years, driven largely by AI agent adoption. Here's the striking stat: 30% of apps running on Vercel's platform are now created by AI agents, not humans. Rauch frames Vercel as the infrastructure layer where all AI-generated software will eventually live. Microsoft is developing its own OpenClaw competitor for Enterprise. The company is testing agent features for Microsoft 365 Copilot with better security controls than the open source OpenClaw project. The idea is an agent that's always working, handling multi-step tasks over extended periods. Unlike OpenClaw, which runs locally on user hardware, Microsoft's version would likely run in the cloud. Expect to see more at their build conference in June. And finally, the largest orbital compute cluster is now open for business. Kepler Communications has about 40 NVIDIA Orin processors spread across 10 satellites, all linked by laser communications. They're serving 18 customers with edge processing for space-based sensor data. The US military is a key customer for applications like missile defense tracking. This is early stage stuff. Real data centers in orbit are still expected to be a 2030s development, but it shows how compute is starting to move to where the data is collected. That's all for today. I'll catch you tomorrow.