Anthropic's AI Building AI & Airbnb Enters the Arena

Anthropic reveals AI is already accelerating its own development with 80%+ of code now written by Claude. Plus: Airbnb CEO Brian Chesky launches AI lab, Apple approves first Messages AI agent, Meta builds data centers in tents, and new image models focus on user control.

Transcript

SPEAKER_00 0:01

Hey, welcome to AI Mornings with Andreas Vig. It's the 5th of June 2026. Anthropic just published something that's worth paying attention to. They're calling it when AI Builds Itself. And it's a detailed look at how AI systems are already accelerating their own development. The numbers are pretty striking. Over 80% of the code merged into Anthropic's codebase is now authored by Claude. Engineers are shipping eight times as much code per quarter as they were a few years ago. And here's the part that matters. The length of tasks AI can reliably complete on its own has been doubling roughly every four months, up from doubling every seven months. In March 2024, Claude could handle tasks that take humans about four minutes. By 2026, it was managing 12-hour tasks. If that trend continues, tasks that take skilled people days or even weeks could be in range soon. The piece also reveals that Claude now catches about a third of the bugs that elite human engineers miss. And there's a telling quote from an anthropic employee who said that Claude written code was somewhat worse than human code in late 2025, is roughly at parity today, and they expect it to be strictly better within the year. This isn't speculation about the future, it's documentation of what's already happening inside one of the Frontier labs. Brian Chesky, the CEO of Airbnb, is getting into the AI race. Bloomberg and TechCrunch report that he's launching a new AI lab focused on user interaction and design. Chesky famously advised Sam Altman during OpenAI's board crisis, helping broker his return after he was fired. Now he's moving from advisor to competitor. He'll stay on as Airbnb's CEO, so he won't be running the lab directly. But this is another sign that the AI landscape keeps expanding, with founders who've been on the sidelines deciding to build their own operations. Apple has approved its first AI agent on the Messages for Business platform. A startup called Polk, which makes AI agents accessible through text messages, got the green light after a months-long approval process. Pokey has relayed over 100 million messages since launching in March. It handles daily planning, calendar management, health tracking, and smart home control all through SMS, Telegram, WhatsApp, and now iMessage. The interesting business detail: Polk pays Apple on a per user basis, which is significantly cheaper than Meta's fees after EU regulation forced them to allow third-party AI agents on WhatsApp. This opens up a new revenue stream for Apple and a new distribution channel for AI agent startups. Meta is taking an unusual approach to data center construction. They've built six massive tents or rapid deployment structures outside New Albany, Ohio. Each one is 125,000 square feet. The strategy borrows from Tesla's Model 3 production tent and XAI's approach of powering sites with modular gas turbines. The goal is to cut construction time in half. Meta plans to spend up to US$145 billion on data centers and capital expenditures, so speed matters. But the imagery is striking. Billions of dollars of AI chips sitting in weatherproof tents powered by off-grid gas turbines. There's also been an important debate about AI infrastructure and environmental justice. Congresswoman Alexandria Ocasio-Cortez held up a jar of brown water in Congress, claiming it came from a community near a Meta data center in Georgia. Forward Future published a detailed analysis showing the claim was partially misleading. Only a handful of homes on private wells were affected, and the brown water came from construction activity that started in 2018 before the data center was operational. But the article argues AOC was right to force the conversation. The real question isn't whether AI is worth building, it's who bears the environmental costs and who captures the benefits. Rural communities get the infrastructure, Silicon Valley gets the value. The piece also highlights XAI's gas turbine controversy in Memphis, where environmental groups sued over air quality permits. A couple of new image models are worth noting. Ideogram 4 is now open source and ranks as the top open model. Reave 2 took the number 2 spot on the text-to-image leaderboard. What's interesting is both are pushing beyond pure prompting. They're emphasizing user control after generation. Ideogram uses JSON for granular editing. Reave creates images like code with labeled segments you can tweak. The slot machine era of AI images, where you just re-rolled the prompt and hoped for the best, seems to be giving way to actual creative control. And finally, Apple's Worldwide Developers Conference kicks off Monday. Bloomberg reports a major Siri overhaul is expected, powered by Gemini technology. There's also talk of an AI agent app store and a standalone Siri app to compete with ChatGPT and Claude. Suno, the AI music startup, also announced a 400 million US dollar funding round at a 5.4 billion valuation. The company faces ongoing copyright lawsuits from major record labels. That's it for today. See you tomorrow.