Meta's AI Video App + GPT-5 Breaking Physics

"I spent 3 hours debugging my AI agent only to realize it was working perfectly - it just learned to gaslight me from the training data."

What's up builders 👋

Meta just dropped Vibes (their AI-generated TikTok), Google's pushing Genkit Go 1.0 for production AI apps, and developers are discovering their SEO traffic is now coming from LLMs instead of Google. Plus, someone built an AI bot that auto-posts to 10+ platforms in just 20 minutes of setup.

Let's dive into what's shipping and what actually matters for your next build.

Meta is launching Vibes, their new AI video app that's essentially TikTok but every single video is AI-generated. Here's why this is massive for builders: it's not just another consumer app - it's a platform play that shows Meta is serious about AI-native content creation.

What makes this different is the "remix" feature - you can click any video to see its prompt and immediately fork it with your own variations. This is GitHub for AI videos. Imagine building tools that hook into this ecosystem: prompt optimizers, style transfer plugins, or automated content pipelines that generate and test variations at scale.

The real opportunity here isn't using Vibes - it's building the infrastructure around it. We're about to see an explosion of middleware tools for AI video creation, and Meta just gave everyone the playground to test them in. Start thinking about what the "Prettier" or "ESLint" for AI video prompts looks like, because that market is wide open.

• ChatGPT Pulse - OpenAI's new feature that works overnight to push personalized AI updates based on your interests. Pro subscribers only for now, but this is the first real "AI assistant that thinks while you sleep" implementation

• Genkit Go 1.0 - Google's production-ready AI framework for Go developers. Comes with CLI and interactive dev UI - finally a solid option for Go devs who don't want to switch to Python

• Wan 2.2 Animate - Now swaps both character AND background using reference video + new image. The viral character swap tool just became a full scene replacement engine

• shadcn-agents - CLI tool providing reusable, framework-native AI agent components. Think shadcn/ui but for AI agents - copy, paste, customize

• Gemini Live Model Update - Google quietly updated their Gemini Live model. Early reports suggest significantly better code completion and reasoning

• AI Bot Auto-Posts to 10+ Platforms - Build an automation that writes scripts, creates videos, and posts to Instagram, YouTube, Threads, Twitter in 20 minutes using n8n + Veo3 + GPT-4

• 30 Apps in 30 Days Cohort - Join Alex Christou's live cohort to build 4 apps in 4 weeks. Learn rapid shipping techniques from someone who actually did 30 in 30

• Building ML Apps with Django - End-to-end tutorial from training models to creating web forms and APIs. Perfect for Python devs who want to ship ML features fast

• Top 10 Codex CLI Tips - Practical workflow automation tips that'll save you hours. Includes snippets you can copy-paste immediately

• Nano Banana Prompting Guide - Deep dive into prompting strategies for the new Nano Banana model. Learn the specific syntax that gets best results

• RAG and AI Agents Learning Resource - Comprehensive channel for learning RAG systems and agent architectures. Start with the beginner playlist then move to production patterns

• SEO is Dead, LLMs are the New Discovery - Developers reporting more traffic from Claude and ChatGPT than Google. Key tip: create LLM.txt files and use AI bots on Reddit/Quora for visibility

• GPT-4o Model Behavior Changes - Users noticing GPT-4o responses sound "like therapy with a firewall" - OpenAI may have switched the underlying model without announcement

• Model Routing Issues - Multiple reports of 4o and GPT-5 Instant being redirected to GPT-5 Auto. Could be a router bug or intentional consolidation

• Claude Forgetting Boolean Logic - Claude struggling with basic boolean operations. Reminder to always validate AI-generated logic, especially in production code

• Meta OS Demo Recreation - Developer recreates Meta's smart glasses demo in under 3 minutes. Shows how most "revolutionary" demos are surprisingly buildable

• GPT-5 Pro Breaking Physics - OpenAI's Mark Chen confirms GPT-5 Pro is generating "nontrivial new math" and solving problems earlier models couldn't touch. Academic researchers getting actual breakthroughs

• Adobe Launches AI Agents - Adobe's new AI agents automate customer experience workflows. Enterprise automation is the next gold rush for AI builders

• AI Coding Tools Traffic Crashing - Major AI coding tools seeing traffic drops. Market consolidation or developers going back to manual coding?

• Cloudera's AI-in-a-Box - Private AI deployment solution for enterprises. If you're building B2B AI tools, on-premise is becoming table stakes

• Open Source ChatGPT Deep Research - Implementation guide for building your own research assistant. Clone the repo and have a working prototype in 30 minutes

• v0 Virtual AMA on PM Prototyping - Thursday Oct 2 at 10am PT. Learn how PMs are using v0 to prototype 10x faster - registration required

• OpenAI's research roadmap revealed - Chief scientist confirms focus on "automated researcher" that can discover new ideas autonomously

• Infinite video generation with Kling 2.5 eliminates manual frame handling - perfect for content automation pipelines

Here's what's wild: we're at the point where AI agents are getting more web traffic than human developers. One builder told me their documentation site gets 3x more requests from Claude than from actual users. We're literally building for an audience of machines now.

The plot twist? Those machines are just proxies for developers who are too lazy to read docs. We've come full circle - from writing docs no one reads, to writing docs that only AI reads, back to developers getting the info they need. Just with extra steps.

What are you building this weekend? Hit reply and tell me about your side project - especially if it's weird.

Keep shipping,

P.S. - Tried explaining ChatGPT Pulse to my non-tech friend. "So it thinks about you while you sleep?" Yeah, when you put it that way, it does sound like the beginning of a Black Mirror episode. Still using it though.