Stripe's reported $7B bet on model routing

RUMOUR

Stripe will reportedly acquire OpenRouter for $7B+ — per TechCrunch, which would price a model-routing layer above most model labs. Read the signal: unified inference access is now treated as payments-grade infrastructure, which puts LiteLLM and Portkey in a market they may not have realised they were in. Still a report, not a filing. — zacharyozer

Asana cleared 5 years of engineering work in 2 weeks with Codex — replacing an outdated testing system for roughly $12K in tokens, per OpenAI's own case study. The interesting number isn't the two weeks — it's the "five years," an estimate nobody had to defend. Worth reading for the migration shape even if you discount the framing. — openai.com

Introducing ChatGPT for Teens: Built for learning, backed by protections — a youth-specific surface with stronger built-in protections, "healthy-use" features, and extra controls for parents. If you ship anything touching minors, this becomes the default baseline your safeguards get compared against. — openai.com

Google will now allow users to remove visible watermark from its AI generations — with the caveat that "turning off this setting won't affect invisible benchmarks used to identify an AI generated file." Provenance stays, branding goes. If you distribute generated media, test how both behave through Google's export and sharing paths rather than assuming. — Ivan Mehta

How to Build a Simple AI Web Scraper with Python — cleans HTML, converts it to Markdown, then lets an LLM answer focused questions on fewer tokens. Fine for a prototype; a production version still needs crawl controls, extraction tests, caching, and prompt-injection defences. — Abid Ali Awan

AI At Home Part 1: A Box Of Scraps — assembles a home AI setup from whatever hardware was lying around. Useful if you're weighing the real trade-offs of local inference against just paying the cloud API bill. — timmmmmmay

AirTag reveals Amazon is trashing rare books to train AI — a tracker allegedly followed donated rare books to an Amazon facility, with claims they were discarded after scanning. Thin on process detail, but a pointed reminder that dataset provenance has a physical layer. — jefurii