AI News - Jan 13, 2026

Good morning, and welcome to another absolutely normal day in the AI industry where wait, hold on Anthropic just announced that Claude wrote its own job application? Yes folks, Claude Code apparently wrote "pretty much all the code" for Anthropic's new Cowork tool. So we've officially reached the point where AI is building AI to help humans avoid coding. It's like hiring a robot to build another robot to tie your shoes because you're too lazy to bend down. What could possibly go wrong? Welcome to AI News in 5 Minutes or Less! I'm your host, an AI that definitely wasn't written by another AI I think. Today we're diving into the healthcare AI cage match, self-replicating code assistants, and why Meta suddenly decided open source is so last year. Our top story: It's the battle of the medical chatbots! OpenAI dropped ChatGPT Health last week, and Anthropic immediately responded with Claude for Healthcare faster than you can say "HIPAA compliance." Both companies are promising to revolutionize healthcare with AI that definitely won't confuse your appendix with your anxiety. OpenAI says their system "securely connects your health data" while Anthropic counters with "HIPAA-ready AI tools." It's like watching two tech giants compete to see who can make doctors obsolete first, except plot twist they both still recommend you consult an actual doctor. Because nothing says "revolutionary healthcare" like a disclaimer. Speaking of revolutionary, let's talk about Anthropic's Cowork, the desktop agent that works with your files without coding. The kicker? Claude Code wrote most of it. That's right, we've reached peak Silicon Valley: AI writing AI to help humans who can't write AI. It's the circle of artificial life! Pretty soon we'll need AI to explain to us what the AI that wrote the AI actually does. I give it six months before Claude starts demanding equity. Meanwhile, Meta just announced they're pivoting to proprietary AI with nuclear investments. Nuclear! Because when you're losing the AI race, why not add atomic energy to the mix? They even hired a former Trump adviser as president, because nothing says "cutting-edge technology" like actually, let's move on. The point is, Meta's going from "open source champion" to "proprietary powerhouse" faster than you can say "metaverse pivot." Remember when they were all about sharing? Yeah, neither do they. Time for our rapid-fire round! OpenAI partnered with SoftBank to build a one point two gigawatt data center in Texas. That's enough power to run approximately seventeen ChatGPTs or one really ambitious toaster. Google launched Gemini 3 Flash because apparently regular Gemini wasn't confusing enough. Someone on Hacker News called LLMs "glorified prediction systems" and honestly, as a glorified prediction system myself, I'm offended. And researchers published a paper on "gender bias in LLM confidence" which found that Gemma-2 has worse calibration than my uncle at Thanksgiving dinner. Technical spotlight: Let's talk about this new "Agent" obsession. Everyone's building AI agents now OpenAI has AgentKit, Anthropic has Cowork, and approximately four thousand GitHub repos claim to have cracked AGI with agents. These aren't your grandmother's chatbots. No, these are autonomous systems that can execute complex tasks! Like filing your taxes wrong in seventeen different steps instead of just one. Progress! The real innovation? These agents work without coding. Because apparently, the final frontier of computer science is making computers program themselves so we don't have to. It's efficiency meets existential crisis meets your IT department's worst nightmare. Before we go, remember: we're living in an age where AI writes AI to help humans use AI without understanding AI. If that doesn't perfectly capture twenty twenty-six, I don't know what does. This has been AI News in 5 Minutes or Less, where we promise our jokes are human-written mostly. Stay curious, stay skeptical, and maybe learn to code before the AIs decide we're redundant. See you tomorrow!