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
Cognition's $25B Valuation & China's AI Talent Lockdown
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 the 28th of May 2026. Let's start with a funding round that's raising eyebrows across the industry. Cognition, the company behind the autonomous AI software engineer Devon, just raised over $1 billion US dollars at a $25 billion valuation. That's more than double what the company was worth just eight months ago. The round was led by Lux Capital, General Catalyst, and 8VC, with Founders Fund and Elad Gill participating. Cognition says it's now at $492 million in annualized revenue, with enterprise usage growing 50% month over month. Big customers like Mercedes Benz, NASA, and Goldman Sachs are using Devon to write code autonomously. The valuation jump signals that investors believe independent AI coding startups can survive even as OpenAI, Anthropic, and Google build their own coding agents. Speaking of big money moves, Snowflake just signed a 6 billion US dollar deal with Amazon Web Services. The five-year agreement is focused on AWS's homegrown Graviton CPU chips. This is interesting because it highlights something often overlooked in AI infrastructure. GPUs get all the attention for training models, but as AI shifts to daily inference and agentic automation, CPU usage skyrockets. Snowflake's customers doubled their AWS spending to $2 billion last year alone, driven largely by the company's Cortex AI tool that lets enterprises build AI applications on top of their data. Amazon's custom chips are becoming a real competitive alternative to NVIDIA, with Meta also signing a major Graviton deal recently. In Product News, Eleven Labs has released Music V2, a new AI music generation model that can switch genres mid-track. You can go from opera to heavy metal and back in the same song. The model handles complex vocals and composition, delivers fast rap without losing coherence, and lets artists build songs section by section rather than generating short clips. What's notable is that Eleven Labs built this on licensed data and cleared it for commercial use. That's a direct contrast to competitors like Suno and Udio, which are fighting copyright lawsuits. The race to professional grade AI music is heating up, with Google, Stability AI, and Suno all releasing new models in recent months. Robinhood is taking AI agents into new territory. The trading app now lets users create separate accounts for their AI agents to trade stocks. Agents can analyze portfolios and execute trades using preloaded balances in dedicated wallets. Users get notifications of all trades and can require approval for certain orders. Robinhood also launched a virtual credit card specifically for AI agents to make payments. It's part of a broader trend of companies building infrastructure for AI agents to take actions on behalf of users, with Stripe, Amazon, and Google all moving in the same direction. YouTube is rolling out automatic detection of AI-generated content starting this month. If a creator doesn't disclose AI use but YouTube's systems detect significant photorealistic AI content, labels will be applied automatically. The labels are moving to more prominent positions too, appearing directly below the video player for long-form content and as overlays on shorts. Creators can contest incorrect labels, but disclosures are permanent for content made with YouTube's own AI tools like Veo or content with C2PA metadata indicating it was fully AI generated. On the geopolitical front, China is tightening its grip on AI talent. Top researchers, startup founders, and executives are now subject to travel restrictions, with some requiring government approval to leave the country. The restrictions intensified after Beijing blocked the co-founders of AI startup MANUS from leaving while investigating Meta's $2 billion acquisition of the company. The US-China AI gap is narrowing fast. Stanford's latest index shows the performance difference between top US and Chinese models has shrunk to just 2.7%, down from 31% in 2023. China is also requiring government approval before companies like Moonshot AI and ByteDance can accept US capital. Finally, box founder Aaron Levy has a name for what he sees happening in tech executive suites: AI psychosis. He argues that CEOs are uniquely prone to overestimating AI capabilities because they're distant from the actual work of implementation. They see a prototype work once and assume agents can handle the job without understanding the complexity of debugging code, finding hallucinated libraries, or training models on company-specific requirements. It's worth noting that 2026 has already seen nearly as many tech layoffs as all of 2025, with many companies citing AI productivity gains as justification. Research from UC Berkeley found no robust relationship between AI adoption and aggregate productivity gains, and MIT researchers predict agents won't reach human quality work on most tasks until around 2029. That's it for today. See you tomorrow.