Shazeer Returns to OpenAI, Microsoft Centers Windows on Agents, Anthropic Hits a $30B Run Rate: Tuesday Briefing, June 23, 2026

Noam Shazeer, co-author of the 2017 “Attention Is All You Need” paper that introduced the Transformer architecture, has joined OpenAI. Shazeer most recently led key efforts on Google’s Gemini project after returning to Google via the acquisition of his startup Character.AI. His departure is both a personal career move and a symbolic one: one of the architects of the foundational technology behind modern AI has chosen OpenAI as the place where he wants to build next.

For OpenAI, the hire adds a researcher whose technical credibility is essentially unmatched in the field. For Google, it represents another in a series of high-profile departures that are eroding the company’s once-dominant position in AI research talent.

At Build 2026, Microsoft made its most ambitious bet yet: agents are becoming the organizational principle of Windows itself. Project Solara and the accompanying MAI-Code-1-Flash model represent Microsoft’s vision of an operating system where AI agents are first-class citizens — not add-ons or assistants, but core components of how users interact with their computers.

The move is bold but carries risk. Previous attempts to embed AI deeply into operating systems — from Clippy to Cortana — failed because the technology was not good enough to justify the intrusion. Microsoft is betting that current-generation models have crossed the capability threshold where always-present AI assistance becomes genuinely useful rather than annoying. If they are right, it could redefine personal computing. If they are wrong, it will join a long list of ambitious Windows features that users immediately disable.

Anthropic has reached approximately $30 billion in annual revenue run rate, a figure that places it among the fastest-growing enterprise software companies ever. The growth is driven primarily by API revenue from enterprise customers, with Claude’s expanding model family and Anthropic’s safety-first positioning resonating with large organizations that need both capability and risk management.

Separately, OpenAI is preparing a confidential IPO filing at a reported valuation of approximately $730 billion. The dual filing dynamic — with Anthropic’s own IPO in the pipeline — sets up a direct market comparison between the two leading frontier labs. Investors will be forced to make explicit bets on which company’s technical approach, business model, and competitive position they believe will endure.

June 2026 is shaping up as one of the densest model-release months in the industry’s history. GPT-5.6, Gemini 3.2, and Claude Fable 5/Mythos 5 are all in various stages of launch from the Western labs. Meanwhile, a tight cluster of Chinese frontier releases is maintaining competitive pressure from the other side of the Pacific. The result is a market where the state of the art is advancing on multiple fronts simultaneously, and the window during which any single model can claim clear superiority is shrinking to weeks.