Here we go again, another week of rapid development in AI. I spent the week at two conferences, one in Berlin and one in Cologne, and also went to one evening event hosted by an AI software company. So I had three days packed with discussions around Agentic AI, LLM, smaller models, datacenters, gigafactories and whatnot.

So this is what I put together for you this week: We start with smaller, specialized language models quietly powering the workhorse AI agents in the background - faster, cheaper, and more nimble at handling real tasks in workflows. Meanwhile, Sierra’s founders Bret Taylor and Clay Bavor are busy reshaping customer service with AI agents that don’t just answer questions but handle complex requests across phones, chats, and apps, essentially acting as personal concierges for customers. On the economic front, AI’s task automation is projected to accelerate rapidly, potentially automating nearly all relevant work within a few years, even though real-world diffusion will vary widely by industry. Behind the scenes, the nonprofit Common Crawl (I had never ever heard of this organization before) has become a controversial lifeline for AI training data, quietly scraping paywalled news articles, raising tough questions about ethics and copyright. And a throwback to AI’s roots: the term artificial general intelligence (AGI), coined in the late 90s, still encapsulates today’s grand ambition for machines that rival human intellect across any domain. Let’s how and when we get there and what this will mean for humanity. Until this happens, I’ll write this newsletter all by myself…

Enjoy this edition of Five Things AI! ...