I’m in Tokyo this week and it’s already Tuesday evening here. Feels like it’ll be a busy week for AI models!
[blog] Our newest TPUs: 8tx and 8i Max, built for the agentic era. One for massive context windows, another for security-sensitive inference. I liked the mini lesson on systems versus silicon.
[article] Anthropic releases Claude Fable 5.2 and Mythos 5.2, pairing new, record-high benchmarks with 20% API cost reduction. Great performance, better pricing, available wherever models are sold.
[article] Meta releases Muse Ember, a new lightweight model for agent execution. Low cost per task in this new Meta model. These aren’t open weights, but competitive on pricing with hosted open-weight models. More here.
[blog] From prompt to app: Build with Google Maps in AI Studio. Maps is a sneaky-rich platform for much more than pushpins on locations. People use these APIs for analytics information, real-time lookups, and more.
[blog] GitHub PRs are slow. Keep shipping anyway. You may experience outages or degradation of cloud services like GitHub, but Adron says you can protect velocity by moving code reviews locally and shrinking change size.
[blog] AI and the Future of the Product Role. Some good thoughts here. The execution-focused product manager/owner is in for a rougher period.
[blog] Google Antigravity for Knowledge Work. Nice touch. Every agent is a coding agent, but we don’t need to throw that interface in front of every type of user.
[blog] A database is a store for global variables. Sure, I guess that’s true. Of course we look at it differently when storing business data. But is it different?
[blog] Bringing conversational analytics to Gemini with Agentic Data Cloud connectors. Speaking of databases, there’s this. I actually just used this feature in a demo where a user could ask for a mix of data. Thanks to these connectors, the agent efficiently figured out what the user wanted and returns it.
[blog] Developers are getting rAIded. None of these are wrong. We have to actively fight against the urge to stop thinking, cede all control, and lose awareness of our code.
[blog] We Benchmarked Coding Agent Harnesses. Thinking Harder Beat All of Them. More agents and more steering (skills, etc) may not be worth a simple increase in thinking level. Sobering.
[article] I Don’t Just Manage Engineers Anymore. I Manage Agents Like a Team. Some of your people-management skills translate directly to agents, some don’t.
[article] Cursor and Google join OpenAI in giving coding agents a voice. Claude Code stays silent. Having ongoing voice interactions with my agent is pretty cool, versus one-and-done recordings. We’ve got a pretty well-designed version in Antigravity.
Want to get this update sent to you every day? Subscribe to my RSS feed or subscribe via email below:
Leave a Reply