Morning, folks. Today's less about what AI can do, more about who controls it, who's winning the business race, whether labs can rein in their own systems, and who actually gets a say as AI takes on more power. Let's get into it.

Today's Top 5

OpenAI is starting to close the gap with Anthropic among U.S. businesses, new Ramp data shows. Anthropic's led since May, hitting nearly 44% by July to OpenAI's 40%. But Ramp's own economist says OpenAI's actually growing faster this quarter so far; developers seem to just like GPT-5.6 Sol.

And Fable 5 isn't pulling its weight either, businesses just aren't picking it up the way Anthropic probably hoped, and the 30-day data retention rule might be part of why. Either way, there's still a month left in the quarter. In this industry, that's more than enough time to flip the whole story again.

Guidelight assessed Anthropic, OpenAI, Google, Meta, and xAI on six basic practices for keeping control of their AI systems: logging, monitoring, gated actions, circuit breakers, third-party reviews and containment plans. None scored above “substantial partial implementation” on any of the six.

Anthropic and OpenAI scored highest overall at C+, followed by Google at D+, xAI at D- and Meta at F. The biggest gaps were around prevention and containment, with companies having less evidence of being able to stop or contain a misbehaving system than detect one.

Meta Just Put AI on Top of Everything You Do on Your Mac

Meta launched a Mac app for Meta AI that can dictate across apps and use what's on your screen to answer questions. Its Muse Spark model can work from whatever you're looking at, whether that's a document, webpage, or another app, so you don't have to keep switching back to a chat window.

Meta is also connecting the app to Instagram, Facebook, Meta Ads, and Google Workspace for business users. The interesting part is less the chatbot and more where it sits: right alongside the work you're already doing.

Anthropic launched Claude Academy on August 20, a free learning hub for people who want to get better at using AI. It covers more than prompts: how to delegate work, check outputs, decide what should stay with you, and use AI without losing important skills.

The interesting part is that Anthropic says these lessons are based on how it approaches AI education internally. The goal isn't just to teach people how to use Claude but how to make better decisions about when and how to use AI.

OpenAI launched AI Futures, a new strategic futures team looking at how advanced AI could change governments, companies, and individual rights. The team says its main question is how free societies can adapt while keeping people in control.

This is less about what the next model can do and more about who gets power when AI becomes capable of doing a lot more. OpenAI is looking at how AI could reshape the way companies and governments operate, much like earlier technologies changed how organizations were built.

Other AI Signals:

  • Anthropic made computer use, the Skills API, and the Files API generally available on Claude Platform, giving developers more ways to build agents that operate software, work through websites, follow company instructions, and create or manage files. In customer testing, one healthcare workflow dropped from 32 minutes to 13 minutes, while the cost per task fell about 30%.

  • Google says its open Gemma models have passed 1 billion downloads, with developers creating more than 100,000 model variants over the past two years. They're now being used everywhere from healthcare and research to satellites and other edge environments, showing how far Google's open models have spread beyond typical AI apps. 

  • Reuters reports that AI is forcing India's $315 billion IT services industry to rethink how it charges clients. Some customers now want the same work for 25%–30% less, while contracts are shifting from paying for hours to paying for measurable results. The Nifty IT index is down about 20% this year, with roughly $73 billion wiped from the combined value of its 10 companies.

  • monday.com rebuilt its platform around AI agents instead of just adding more AI features. More than 250,000 companies use monday.com, and users have already had 5 million+ interactions with AI agents since May. The bigger change is simple: people and agents now work inside the same projects and workflows.

  • Broadcom is reportedly in talks to raise more than $60 billion in debt for an AI chip financing deal supporting Anthropic and other AI companies. Bloomberg says the package could eventually reach $100 billion, although nothing has been finalized. The scale shows just how much money is now needed to build the computing infrastructure behind the AI boom.

Today's AI Tools to Try:

  • Simple Pages: Quickly build clean, conversion-focused landing pages without the usual complexity.

  • AgiLoop: Reviews your codebase for hidden risks and weak spots, then turns its findings into production-ready improvements.

  • Tegaki: Turns any font into realistic animated handwriting using generated stroke data, with support for popular JavaScript frameworks and plain JavaScript.

  • Skylum: AI photo editing software for enhancing images, retouching portraits, replacing skies, removing backgrounds, and more with simple controls.