Today's big theme is personal agents: Anthropic launched Claude Duet, an agent that comes to you first, and Meta told investors that Muse has passed 10 million users. Meanwhile, OpenAI is bringing its models to Google Cloud, two AI startups closed major rounds, and Recursive open-sourced a coding model that runs on a laptop.

1. Anthropic launches Claude Duet, a proactive agent app

On October 21, Anthropic opened a research preview of Claude Duet, starting with US Pro and Max subscribers via a waitlist. Claude Duet is a standalone app (iOS, Android and claude.ai) built around Claude Opus 5.5. Each morning it sends a short brief on the most important items across your connected apps, checks them every few hours during the day, and messages you only when something needs you. It has its own virtual machine and browser, so work continues after you step away. Before it sends a message, makes a payment, signs a document or deletes anything, it shows exactly what it will do and waits for approval; some actions, such as trading securities or moving money between accounts, are off-limits even with approval. It can use more than 6,000 connectors and plugins from the Claude Marketplace, including Gmail, Slack, QuickBooks, Stripe, Gusto, Docusign and Shopify.

Why it matters: Following Meta's Muse, Anthropic is moving agents from "you ask, it answers" to "it comes to you," with approvals and sensitive-data safeguards as selling points.

Source: Claude

2. OpenAI's frontier models and Codex are coming to Google Cloud

OpenAI announced on October 21 that it is partnering with Google Cloud to bring its frontier models and Codex agentic workflows into Google Cloud's Model Garden. Customers will be able to buy them through Google Cloud Marketplace and draw down existing GCP spend commitments, and OpenAI models will soon be available in Google Antigravity and the Gemini Agent. General availability is planned in the coming weeks. OpenAI noted that Microsoft remains its primary cloud partner.

Why it matters: Two direct model competitors are cooperating on distribution, giving enterprises more model choice on one cloud.

Source: OpenAI

3. Meta: Muse passes 10 million users, with shopping a top use

On its third-quarter earnings call on October 21, Meta said its AI agent app Muse has more than 10 million users. Mark Zuckerberg said about 25% of people who set up Muse now use it daily, and more than a quarter of engaged users have shopped through it, with an average purchase above $200. Meta plans to keep Muse free for most users and take a share of commerce on the platform, and expects to improve Muse's compute cost efficiency by more than 10x over the next year.

Source: TechCrunch

4. Funding: Flapping Airplanes raises $1B, Legora valued at $8.8B

Year-old AI lab Flapping Airplanes announced a $1 billion Series A co-led by Index and Lux, reportedly at a $6 billion valuation, four times the $1.5 billion valuation of its seed round nine months ago. The company aims to train large models with far less data and has built a "megakernel" framework that can run training and inference inside a single GPU kernel. The same day, legal AI company Legora announced a $550 million Series E at an $8.8 billion valuation, led by Atreides Management; it said ARR now exceeds $200 million and the money will fund expansion in North America.

Sources: TechCrunch on Flapping Airplanes; TechCrunch on Legora

5. Open source: Recursive releases Flint, a coding model for laptops

Recursive, led by Richard Socher, released its first model, Flint, as open weights on Hugging Face under the Apache 2.0 license. Flint is a mixture-of-experts model with about 40 billion parameters, of which about 2.5 billion are active per token, so it can run locally on a laptop. In the company's tests it generated about 131 tokens per second on an M4 Pro MacBook Pro and outperformed other models under 50 billion parameters on SWE-Bench Verified. Recursive says its automated research system, the Eureka Machine, contributed to nearly every step from pre-training to inference, letting a small team build the model from scratch in under four months.

Why it matters: It is a public test of AI helping to build AI, and a lightweight new option for private, offline coding.

Sources: TechCrunch; Hugging Face: RecursiveSI/flint

6. OpenAI coordinates fixes for critical PHP vulnerabilities

OpenAI disclosed on October 21 that over the past two months it led a coalition of maintainers, researchers and security providers to address critical vulnerabilities in PHP. Security firm Calif found one of the flaws using GPT-5.6 Sol through OpenAI's Patch the Planet initiative, Wiz independently discovered and reported both, and PHP's maintainers led the fixes and release. OpenAI called it one of the most impactful security issues in recent memory, urged PHP users to apply the fixes, and shared a PHP version checker as a Codex skill. It also committed $1 billion in subsidized access for defenders of essential services with limited security resources.

Source: OpenAI

---

In one line: Personal agents are becoming proactive and reaching scale, frontier models are spreading across clouds, and investors keep backing more efficient training and vertical AI.