TL;DR

• Anthropic launches Claude Duet, an AI agent that works on its own.
• OpenAI brings its models and Codex to Google Cloud.
• SpaceX posts a record quarter, and Elon Musk floats a merger with Tesla.

Anthropic launches Claude Duet, an AI agent that reaches out to you first

Anthropic, the maker of Claude, launched Claude Duet as a research preview on October 22. Duet is an AI agent: a program that doesn't just answer but completes tasks on its own. It comes as a standalone app for iOS, Android and claude.ai. At launch, access is by waitlist for Pro and Max customers in the US. It runs on the Claude Opus 5.5 model.

Duet doesn't wait for a question. Every morning it sends a briefing drawn from your connected apps; more than 6,000 integrations are available, including Gmail, Slack and Stripe. It keeps working on its own cloud computer, even when your laptop is off. Payments and outgoing messages need your approval. Duet refuses to trade securities or move money between accounts.

Why it matters: Competition is shifting from the chat window to agents that work in the background. TechCrunch names Meta Muse and OpenAI Dots as direct rivals. On the same day, Anthropic handed its open standard Agent Skills to the Linux Foundation's Agentic AI Foundation. Agent Skills lets agents learn specialized knowledge. According to Anthropic, more than 70 platforms use it, including ChatGPT, Cursor and GitHub Copilot.

• Claude Duet app is Anthropic's bet on agents
• Anthropic takes on Muse with Claude Duet, its new AI agent
• Donating Agent Skills to the Agentic AI Foundation

OpenAI brings its models and Codex to Google Cloud

OpenAI is bringing its frontier models and its coding agent Codex to Model Garden, Google Cloud's model catalog. Google's coding tool Antigravity follows soon. Companies can pay with their existing Google Cloud budgets. General availability follows in the coming weeks. For context: cloud providers such as Google, Amazon and Microsoft rent out computing power and sell AI models to businesses.

Why it matters: According to CNBC, this puts OpenAI on all three major US clouds. It follows the end of Microsoft's exclusive license. OpenAI says Microsoft remains its most important cloud partner. For Codex, the deal means more distribution right where developers already work, including on the home turf of Google's own Gemini models.

• OpenAI to bring AI models to Google's cloud, in further expansion beyond Microsoft
• OpenAI models are coming to Google Cloud

Cognition opens Devin Oncall to all customers: its AI agent takes on on-call duty

Cognition builds Devin, an AI software engineer. The company also took over Windsurf, now called Devin Desktop. Since October 19, Devin Oncall has been available to all customers. The agent watches alerts from Slack, Teams and PagerDuty and triages incidents. It digs through code, deployments and monitoring to find the likely cause. In a video interview with Bain Capital Ventures, CEO Scott Wu says Devin now writes more than 95 percent of Devin's own code.

Why it matters: Many agents can write code by now. The fight is moving to everything that comes after: review, shipping and incident response.

• Devin release notes 2026 (Devin Oncall GA)
• Video: Cognition on Building the AI Software Engineer

AlphaGo co-creator Thore Graepel launches AI startup Metis Reasoning in London

Thore Graepel was a head of research at Google DeepMind and a co-developer of AlphaGo. That program beat the world's best Go players. Graepel left DeepMind in the summer. On October 20 in London, he unveiled Metis Reasoning. The twelve-person team includes co-founders Piers Ingram, Michael Warren and David Berman. According to Tech.eu, Metis initially aims to raise a sum in the tens of millions.

Why it matters: Metis isn't betting on ever-larger language models. Instead, it pursues traceable, compute-efficient reasoning modeled on AlphaGo. That makes London a hub for prominent alternatives to OpenAI and Anthropic, alongside David Silver's Ineffable Intelligence and Richard Socher's Recursive Superintelligence.

• Former Google DeepMind top researcher launches London compute-efficient AI company

OpenAI hires energy and partner sales specialists from AWS

OpenAI has hired Will Friedman as Energy Regulatory Lead. He was most recently Senior Corporate Counsel at Amazon Web Services (AWS). At OpenAI, he will coordinate permits and power deals for new data centers. Ignacio Guillen joins as Go-to-Market Partner Director. He was previously a partner sales leader at AWS.

Why it matters: Power is the bottleneck in the AI buildout. Enterprise customers are often won through partners. For both, OpenAI is recruiting directly from the cloud giants. It is their customer and their competitor at the same time.

• OpenAI hires new energy regulatory lead
• OpenAI Hires Former AWS Partner Sales Leader To Drive AI Channel Growth

SpaceX posts record quarter, Elon Musk floats merger with Tesla

SpaceX is now publicly traded. Besides rockets and Starlink, it now also runs an AI business. In the third quarter, revenue rose 161 percent to $14.58 billion, against expectations of $12.72 billion. Net income came in at $4.4 billion, after a loss of $1.05 billion a year earlier. According to CNBC, the AI business grew ninefold.

Why it matters: On the analyst call, Elon Musk said Tesla and SpaceX have formed special committees of independent directors. They will examine a merger, which Musk calls a good idea. There are no terms yet. On Tesla's side, a majority of all shares would have to approve. According to CNBC, SpaceX stock jumped about 10 percent.

• Tesla and SpaceX are discussing a possible merger, Musk says
• SpaceX pops 10% after revenue rose 161%, Musk floats Tesla merger

AI finds critical PHP vulnerabilities, OpenAI coordinates the fixes through Patch the Planet

PHP is a programming language that runs a large share of websites. Security firm Calif used OpenAI's GPT-5.6 Sol model to find a critical vulnerability. Wiz discovered the flaws independently. Through its Patch the Planet initiative, OpenAI coordinated fixes with maintainers, Cloudflare and government agencies before disclosure.

Why it matters: OpenAI calls it one of the most consequential security issues in a long time. It expects AI to greatly speed up findings like this, for defenders and attackers alike. OpenAI is also committing $1 billion in subsidized access to Daybreak, its security offering, for under-resourced defenders.

• Patch the Planet: global response coordination as agents uncover critical vulnerabilities

Flapping Airplanes raises $1 billion, reportedly at a $6 billion valuation

Flapping Airplanes is a year-old AI lab that wants to train large models on far less data. It raised a $1 billion Series A led by Index and Lux. Reports put the valuation at $6 billion. That comes nine months after a $180 million seed round at a $1.5 billion valuation.

Why it matters: The valuation has quadrupled in nine months. Investors such as Sequoia, GV and Nvidia are betting that efficiency will be the next big race: more intelligence per chip and per dataset.

• Year-old AI lab Flapping Airplanes raises $1B at a $6B valuation

20VC episode 'Trillion or Bust': why megafunds only bet on trillion-dollar exits

In the October 22 episode, Harry Stebbings, Rory O'Driscoll and Jason Lemkin discuss why megafunds now only bet on trillion-dollar exits. Solid companies in between can barely raise money. Topics include talks about a lower valuation for Anthropic's IPO (about $1.5 trillion instead of $2 trillion, according to Reuters), Modal's $750 million round at $15.75 billion and Anduril's $90 billion tender.

Why listen: Lemkin calls code review the most important category in software right now. That fits the Cognition story in this edition.

• 20VC x SaaStr: $20M ARR and Unfundable, Anthropic's $16.8B Quarter, and Modal at $15.75B
• Investors temper Anthropic IPO valuation expectations