AI News · The field notes
The Daily AI Drop - Evening Edition, November 10, 2026
Fable 5.2 gets two opposite verdicts in one day, OpenAI puts businesses inside ChatGPT, and the agent era moves from promise to plumbing.
Thirteen new videos landed this afternoon, and most of them point the same way: AI is moving out of the demo and into real work. Mehul Mohan gave Claude Fable 5.2 two opposite verdicts in a single day, OpenAI turned ChatGPT into a place where businesses live, and the agent era's unglamorous plumbing — memory, testing, grounding — took center stage. Underneath it all, the Nvidia chip fight with China kept escalating.
Listen: Listen to this episode
NEW Claude Fable 5.2 Beats EVERYTHING — mehulmpt
Mehul's second Fable 5.2 video of the day is the bullish one: the new Claude model beats everything he throws at it. It matters because it is the other half of a same-day split verdict — hype and skepticism arriving hours apart — and the gap between the two is where the real evaluation of a flagship model happens.
Fable 5.2 Review (Don't use it) — mehulmpt
Mehul's first Fable 5.2 video of the day is the cautionary one, arguing against using the model despite the launch buzz. Paired with his later "beats everything" take, it captures the first wave of flagship-model reactions perfectly: the truth usually shows up a week later, once people have actually built something with it.
Which Free AI Tools Can Actually Build a Business? A VC Decides — TheNextNewThingAI
A venture capitalist judges seven open-source AI projects on whether they can actually carry a real business, separating tools that can support a company from ones that are just fun demos. If you are picking a stack to build on, this is the filter that matters: viability, not novelty.
Try Matt Pocock's 'Grill Me' Skill Before Your Next AI Project — TheNextNewThingAI
Matt Pocock's Grill Me skill puts your AI project plan through a tough interrogation before you build it. Letting the model poke holes in your plan while it is still cheap beats discovering them after you have shipped — a small practice with an outsized payoff.
Claude-Mem Gives Your AI Agent a Memory You Can Control — TheNextNewThingAI
Claude-Mem adds a memory layer to AI agents that you control. Agents that forget everything between sessions are barely useful, and agents that remember everything are a privacy problem — controllable memory is the middle ground that makes agents genuinely usable day to day.
China Won't Take 600,000 Nvidia Chips. Did It Target Jensen Huang For Better Ones? — NateBJones
Nate walks through the reporting on China's refusal of six hundred thousand Nvidia chips and asks whether Beijing targeted Jensen Huang to get better ones. The chip standoff is now about who gets the hardware that trains the next generation of models, landing right as AI starts helping build the next AI — a supply-chain story for anyone who depends on AI infrastructure.
Answering your questions about GPT-Live — OpenAI
OpenAI answers the internet's questions about GPT-Live: how it differs from earlier voice models and what developers are already building with it. Voice is becoming a real interface for agents rather than a party trick, and OpenAI clearly wants developers to treat it that way.
Introducing Businesses in ChatGPT in beta — OpenAI
Businesses in ChatGPT is now in beta: a business can claim its listing, manage its information, and see how people engage with it inside ChatGPT. If people start finding and talking to businesses inside the chatbot instead of through a search engine, showing up there becomes as important as having a website.
Greg Brockman on how OpenAI uses frontier models for cyber defense — OpenAI
OpenAI president Greg Brockman shares how OpenAI uses its own frontier models for cyber defense, turning lessons from defending OpenAI into tools for customers. The same models that can write an attack can also watch for one — and the companies building the models are starting to sell the shield, not just the sword.
How CodeRabbit uses OpenAI to review AI-generated code — OpenAI
CodeRabbit co-founder and CEO Harjot Gill explains how his team uses OpenAI models to review code produced by AI agents and decide what is ready to ship. Agents write the code, models review the code, and a human still makes the final call — that review step is where a lot of teams are going to live from now on.
How Workday Builds Agent Prototypes in Under a Day with LangChain — LangChain
LangChain shows how Workday builds agent prototypes in under a day. A company of Workday's size prototyping agents in hours means the bottleneck is no longer the technology — it is knowing what to build.
Microsoft Foundry: Test Your Agent Before You Ship — MicrosoftDeveloper
Adrian Twarog tests an agent in the Microsoft Foundry playground before publishing it, checking it with a single prompt. Testing before shipping is the discipline that separates a demo from a product — and it is the part of agent building most teams skip.
Grounding Agents with Real World Intelligence — MicrosoftDeveloper
Microsoft Web IQ gives AI agents fresh, trustworthy information from across the web, with grounding, passage-level retrieval, citations, and scopes. Ground the agent so it is not making things up, then test it before it touches a real user — that is the whole discipline in miniature.