Daily Brief

OpenAI spun out DeployCo, a new business unit focused on helping enterprises move frontier AI from prototype to production. The unit targets the operational gap between API access and working business systems—the exact layer where most enterprise deployments stall. No pricing or staffing numbers have been disclosed.

Airbnb reports 60% of new code is AI-generated via Claude, with CEO Brian Chesky noting that even non-engineering managers are now writing code directly. This follows Shopify at 50% and Google at 75%, establishing a clear industry benchmark pattern across major tech companies.

Anthropic's Claude Platform launched into general availability on AWS, bundling managed agents, code execution, web search, and batch processing. Critically, Anthropic committed to same-day feature parity with its native API—meaning AWS customers won't lag behind direct API users on new capabilities.

An unnamed Fields Medal-winning mathematician publicly stated that GPT-5.5 is now solving open mathematical problems at PhD thesis level, warning the math research community faces an imminent crisis of relevance. This is not a benchmark claim—it's a domain expert asserting that novel research-grade problems are being solved.

Hermes Agent overtook Claude Code and OpenClaw to become the highest token-consuming model on OpenRouter's global metrics over a 24-hour window. The surge reflects concentrated agent workload traffic rather than broad consumer adoption.

Ben Thompson's Stratechery argues that agentic inference operates on entirely different economics than interactive inference—when no human is waiting, latency optimization becomes irrelevant and throughput, cost-per-token, and parallelism dominate. The piece reframes the entire inference infrastructure buildout as solving the wrong problem for the agentic era.