Today's briefing

What matters today

Today’s throughline is AI at the handoff: Stripe is reportedly pursuing a deal that would bring model routing closer to token billing, while new systems reshape how work, discovery, and oversight move across the internet. The details matter—from routing neutrality and private safety signals to meeting assignments and deceptive ad creative.

Lead story / acquisition

Stripe’s OpenRouter Deal Puts AI Routing Beside the Billing Stack

The reported Stripe deal for OpenRouter is expected to close within weeks and would bring an AI model-routing layer into the company that already supplies OpenRouter’s payments infrastructure and helped build token billing. Neither company has confirmed an agreement or disclosed terms: published estimates range from $7.5 billion to more than $8 billion, with Axios reporting consideration that was mostly stock.

OpenRouter offers developers a single endpoint across more than 400 AI models from more than 80 providers. It says more than 10 million developers and companies use the platform and that it handles more than 10 trillion tokens daily; its system scores requests by complexity, price, and speed, letting developers change providers without changing code.

That makes routing more than an integration convenience. Stripe already sits alongside the payment and metering flow, while OpenRouter determines where a request goes; together, those layers connect a model choice and provider selection with the spending they generate. One account reported OpenRouter retains about 5% of inference spending processed on the platform, though transaction terms remain undisclosed.

OpenRouter says its product, mission, commitments, and customer-first routing will remain unchanged after closing. The practical test comes after close: whether its routing decisions continue to reflect what is best for customers while its new owner gains a deeper position around AI usage and billing.

platform shift Anthropic’s Meeting-to-Agent Blueprint Is Still Only a Blueprint

Reverse-engineered Claude Desktop artifacts describe Parka, a workflow for turning recorded meetings into speaker-attributed transcripts, editable notes, and assignments for Claude Cowork, Claude Code, or people. Public builds expose no usable interface, so release timing, branding, and automatic execution remain unresolved.

security risk Meta’s Ad Review Missed Sexualized Deepfake Promotions

Meta ran 32 ads for Kromix, an AI image tool that promoted unrestricted sexualized content involving real people. Most used benign cover images to conceal pornographic material; Meta removed them after WIRED asked questions, while Apple also removed Kromix from its App Store.

enterprise adoption One School Makes AI Permission an Assignment Decision

Washington Leadership Academy lets teachers set AI use from zero to unlimited for each assignment, alongside instruction in prompting, output evaluation, and tool selection. Its classroom results are promising but limited: a Stanford review found gains often disappeared once AI access was removed.

A lighter look at a consequential AI platform story.

Reader check-in

Help shape tomorrow's briefing

One click tells us what to keep, improve, or tighten.