Thursday, August 20, 2026

Share this briefWhatsAppLinkedInX

The Brief

The heaviest crown in tech just changed hands. Recent reports indicate that Anthropic has surpassed OpenAI in revenue for the first time. This milestone validates the massive growth we noted earlier this week when Anthropic hit a $65 billion annualized run rate (August 18 edition). The corporate rivalry is actively shifting market share, and enterprise adoption is clearly moving toward Claude.

OpenAI is working to secure its enterprise flank in response. The company just formalized a zero data retention policy for eligible API customers using its frontier models. If you process sensitive corporate data, OpenAI guarantees it will not store your information or use it for training. It is a necessary table stake for businesses wary of leaking trade secrets to a language model.

Getting enterprise trust is one thing; getting into your daily workflow is another. Meta just released a dedicated Mac app for Meta AI that features screen sharing and dictation. By letting the assistant see your screen, Meta provides context-aware suggestions directly on your desktop. This closely mirrors OpenAI's recent push to log screen activity and keystrokes locally (August 16 edition). AI developers realize that if they want to be useful, they need to live outside the browser tab.

Even your calendar wants in on the context. Calendly just launched an AI note-taker and scheduling assistant named Callie. The meeting automation space is already crowded, but bundling pre-meeting prep and post-meeting documentation into the tool you actually use to book the time makes practical sense for a fragmented workflow.

You can do a lot more with that context than just summarize meetings. AI researcher Andrej Karpathy recently detailed how he replaced traditional note-taking by using an LLM to structure and query a personal wiki. Maintaining a searchable repository of learnings creates a living database with almost zero friction. If you want to take it a step further and build custom tools on top of that data, Replit just launched a free mode utilizing OpenAI's GPT-5.6 Luna. You can prototype software using natural language without worrying about token costs. As AI agents reshape software engineering skills, the line between writing a note and building an application is fading fast.

The infrastructure to power all this reasoning requires deep pockets. Financial giants like BlackRock and Goldman Sachs are partnering with Nvidia to create $500 billion in financing. They are officially treating GPU compute as an asset class, ensuring the hardware powering these models remains heavily funded while the software layer fights for dominance.

Bottom line: The operating system of your daily work is changing. With Meta and OpenAI fighting for access to your Mac screen, AI is shifting from a tool you query to an assistant that observes your context. Use this moment to rethink how you store information. Structuring a personal knowledge base for an LLM to read will soon be far more valuable than manually typing out meeting notes.

The details

The AI race

5 items
  1. OpenAI introduces zero data retention for frontier models via API

    OpenAI is formalizing its commitment to data privacy by offering Zero Data Retention for eligible API customers using its most advanced models. This allows enterprises to process sensitive data through frontier models without the risk of their information being stored or used for training.

  2. Anthropic reportedly surpasses OpenAI in revenue for the first time

    Recent reports indicate a significant shift in the competitive landscape as Anthropic's revenue has overtaken OpenAI's. This milestone highlights the intensifying rivalry and shifting market share between the two primary developers of frontier AI models.

  3. Financial giants and Nvidia partner to turn compute into an asset class

    Major investment firms like BlackRock and Goldman Sachs are creating $500 billion in financing to treat GPU compute as a financial asset. This shift in capital structure ensures that the underlying hardware powering AI models will remain heavily funded and accessible for enterprise scaling.

  4. China allows Nvidia H200 chips onto mainland to bolster AI development

    Small quantities of Nvidia's high-end H200 chips are reportedly reaching Chinese AI firms despite export restrictions. This supply shift is critical for global AI competition as it allows Chinese labs to continue training advanced models that compete with US-based frontier labs.

  5. GLM-5.3 ties for top open model spot with aggressive pricing

    Chinese startup Z.ai has announced GLM-5.3, which matches the highest performance scores currently held by open-weight models. Knowledge workers should watch this space as the model aims to significantly undercut competitors on price, further driving down the cost of high-intelligence reasoning.

AI at work

4 items
  1. How an LLM-powered personal wiki transforms professional knowledge management

    AI researcher Andrej Karpathy details how using LLMs to structure and query a personal wiki has replaced traditional note-taking. This approach allows knowledge workers to maintain a searchable, living repository of their thoughts and learnings with minimal friction.

  2. Calendly launches AI meeting note-taker and scheduling assistant Callie

    Calendly is entering the competitive meeting automation space with a new AI note-taker and an assistant named Callie. Knowledge workers can use these features to streamline pre-meeting preparation and post-meeting documentation directly within their scheduling workflow.

  3. Meta AI launches Mac app with screen sharing and dictation features

    Meta is releasing a dedicated macOS app that can 'see' your screen to provide context-aware suggestions and content generation. This moves Meta AI closer to the system-level integration seen in ChatGPT and Copilot, making it a more viable assistant for daily desktop workflows.

  4. Amazon makes upgraded AI Alexa+ free for Fire TV users

    Amazon is rolling out an AI-enhanced version of Alexa to all Fire TV users in the US without requiring a Prime subscription. This broadens access to conversational AI interfaces in consumer hardware, though it is currently focused on media consumption.

For builders

2 items
  1. Replit launches free software creation powered by GPT-5.6 Luna

    Replit has introduced a Free Mode that utilizes OpenAI's GPT-5.6 Luna model, removing token cost barriers for building applications. This allows non-technical users and developers alike to prototype software via natural language with high-reasoning capabilities.

  2. Liquid AI releases LFM 2.5 checkpoints using quantization-aware distillation

    Liquid AI has released new optimized checkpoints for their Liquid Foundation Models (LFM) 2.5, focusing on efficient 4-bit quantization. This is relevant for developers looking to deploy high-performance models on edge devices or hardware with limited memory without losing significant accuracy.

    Hugging Face BlogRead the full article

Hands-on

1 pick
  1. Tutorial

    Addy Osmani on how AI agents are reshaping software engineering skills

    A deep dive into how AI agents are fundamentally changing developer workflows and the specific skills engineers need to stay relevant. It provides high-level insights for knowledge workers on the transition from traditional coding to AI-augmented engineering.

    The Pragmatic EngineerRead

Get tomorrow's brief, curated for you.

Subscribe →