AI Field Notes: Ex-Google, DeepMind, and more

“An AI video of a woman walking down a street with famous meme characters has racked up 2 million likes on Instagram.”

“Anthropic published a blog post on the best practices for building with Claude Sonnet 5.5. It primarily breaks down when to use Sonnet over Opus.”

“You can also reply directly to this email if you have suggestions, feedback, or questions.”

“This guide breaks Claude Code spend down by the things that really move the bill: turns, cache reads, output tokens, effort level, and model choice. You’ll also see when to use medium vs. high effort, how to keep the cache warm, when /compactpays off, and how to use /usage to measure the real cost of your own tasks.”

“--------View image: (https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/18efc0eb-c3c4-483f-a001-0fe0dcca16c3/GroupfromFigma2.png?t=1758120539)”

“--------View image: (https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/18efc0eb-c3c4-483f-a001-0fe0dcca16c3/GroupfromFigma2.png?t=1758120539)”

“Create an actionable (and defensible) AI budget”

“The HUMAN Guide to AI and Agentic Traffic shows CISOs how to:”

“Team Bots let you build an agent around a specific role or workflow, then share it with teammates. The Bot builds two memories over time that are kept separate: context of individual conversations and shared team skills. It’s available now on Teams and Enterprise plans. Separately, Grok Bot can now manage your finances”

“Anthropic’s Claude Sonnet 5.5 is now available, offering performance that tops Fable 5.1 on one index, while running 30% faster and 30% cheaper than Sonnet 5 (here’s the prompting guide). Separately, ElevenLabs rolled out Eleven v4, its most emotive voice model yet, built on a new architecture to improve tone, pacing,”

“it loops in human assistants to complete tasks AI can’t do on its own. The startup claims this strategy helps it outperform competitors on trust and task completion rates, although its benchmark has not been independently verified. Watch how it works.”