AI in a Nutshell - Week 51 - AI Agents Build Their Own Society, Claude Hits Chrome, and ChatGPT Opens App Store
Claude ships a Chrome extension, you can now pin chats in ChatGPT, Google lets you verify AI videos in Gemini, and developers can now submit apps to ChatGPT.
Fellow human, it’s week 51. AI hasn’t taken over the world yet. And here’s your lazy man’s version of what went down in the AI world in the past week.
Summary for extra-lazy people: 1,000 AI agents built their own Minecraft civilisation, Claude launched as a Chrome extension, you can now pin chats in ChatGPT, Google lets you verify AI videos in Gemini, and developers can submit apps to ChatGPT.
What You Must Know
You Can Now Pin Chats in ChatGPT → ChatGPT now lets you pin chats to the top of your list. It works on iOS, Android, Web, and Atlas.
1,000 AI Agents Left to Build Their Own Village → So this company ran an experiment where they dropped 1,000 AI agents into Minecraft and just let them do their thing for days. They built farms, markets, governments, and even religions. Some became leaders, others priests, and a few got corrupt. It was weirdly human, complete with cliques and gossip. The whole point was to see if AI agents could actually work together, but turns out they were too independent. Most users just wanted them to follow orders.
Claude is now on Chrome as an Extension → Claude now has a Chrome extension. It can see your screen, click buttons, fill forms, basically use the browser as you would. In theory, I find it similar to ChatGPT Atlas, just less capable because it’s just a Chrome extension.
You Can Now Verify Google AI-Generated Videos in Gemini → Google added a feature to Gemini where you upload a video and ask if it’s AI-generated. It scans for SynthID watermarks and tells you which parts are AI. The issue is that it only works for videos generated with Google’s AI. I tested it with a Sora video, and it couldn’t detect anything because Sora doesn’t use Google’s watermarks. I don’t know if it already exists, but a tool that tells whether any content is AI-generated (with great accuracy) will have a good use case in this age.
Developers Can Now Submit Apps to ChatGPT → OpenAI opened app submissions for ChatGPT. There’s now an app directory at chatgpt.com/apps where developers can submit apps using the Apps SDK.
What’s Good to Know
OpenAI Adds Teen Safety Rules to Model Spec → OpenAI updated ChatGPT’s rules for users under 18. Teen safety now comes first, even over being helpful. My first thought was, how would they know who is what age?
EU Drafts Rules to Label AI-Generated Content → The EU published a draft to mark AI-generated content with watermarks and labels. It’s part of the AI Act, which requires AI providers to mark synthetic stuff and deployers to label deepfakes. I believe some providers already do this e.g Google.
OpenAI Ships GPT-5.2-Codex for Long-Context Coding → OpenAI released GPT-5.2-Codex for long coding sessions without losing context. It’s optimized for refactors, migrations, and big feature builds. I’ve never been a fan of CODEX cos my first experience with it was annoying, but I’ll keep this in mind.
Anthropic Makes Agent Skills an Open Standard → Anthropic released Agent Skills as an open standard, as they did with MCP. ICYDK, Skills are folders containing instructions that tell the AI how to perform specific tasks. Microsoft already adopted it in VS Code and GitHub. OpenAI quietly copied the same structure in ChatGPT and Codex CLI. Anthropic just gave it away for the culture.
Money Moves This Week
💰 Neurable raised $35 million to advance its AI-powered brain-computer interface technology.
💰 CuePilot AI raised $1.8 million to enhance its AI capabilities and expand globally.
💰 Mythic Inc. raised $125 million to challenge NVIDIA.
💰 NobodyWho raised €2 million to power on-device AI models.
💰 FINNY raised $17 million to help great advisors help more people.
Kaizen Corner → How to Self-Host n8n if You’re a Lazy Person
Using n8n extensively on its cloud option is kinda expensive compared to hosting it yourself, and that’s aside from data privacy.
Now, there are a lot of ways to self-host n8n, but I always tend to go with the laziest way, which takes less than 5 minutes if you don’t have shitty internet.
The approach is: Using render.com and the n8n Docker image. Just follow this guide: https://render.com/docs/deploy-n8n
Render handles the deployment for you, and you’re up and running without dealing with server configs or terminal commands. It’s the easiest path if you just want to get n8n working without the usual DevOps headache.
Meme of the Week
That’s your week in AI.
If you learned something, tell a friend. And if you didn’t, blame yourself :)
Until next Sunday,
Kay - your fellow human.
P.S. If this email lands in spam, that’s your inbox trying to stop you from staying plugged in. Fix it.
Quick Note: I’m not an expert, I’m just a curious normie navigating the world of AI (like you). Every Sunday morning, I wake up super early (before the kid wakes up), go online, and read about AI, like my dad used to read newspapers back in the day. Then I write down everything new I discovered and turn it into this newsletter for anyone else who’s curious enough to read it. Why? Because I learn faster and understand deeper when I have to explain it in writing. I’m welcome.