Good morning. It's Tuesday, November 10, and Fable 5.2 might be the best AI model we've ever used. We put it through five demanding real-world tests, including building fully functional alternatives to Excel, Photoshop, and Premiere. The results convinced us we're entering a new era of software development. But there's a catch, and it's a big one. More below.

Plus today's top stories: NVIDIA's ties complicate a chip-smuggling probe, Ukraine puts AI to work on the battlefield, and Google doubles down on specialized AI chips.

→ Make Enterprise Content More Useful at Every Step

Fable 5.2 is the best AI model I've ever used, by far.

After spending days pushing it to its limits, I'm convinced we're entering a completely different era of software development.

I gave Fable a three-line prompt and got a playable Gears of War clone with incredible graphics, enemy AI and real combat mechanics. I asked for Microsoft Excel and got the most feature-complete spreadsheet clone I've ever built. My team created everything from professional video editors to multiplayer games, custom Android utilities and complex 3D environments.

These were working applications, not mockups, and some are already being used in our actual production workflows.

I've built versions of some of these projects with earlier models, and nothing came close to this level of completeness. That's why I think Fable 5.2 represents a meaningful leap, not just another incremental upgrade.

But how much of this translates into practical, everyday development? And when does the cost outweigh the benefit?

→ Explore all five field tests, including the results, costs, and lessons learned

Google introduced TPU 8i Max for models that need very large context windows and TPU 8tx for confidential AI inference. A TPU (tensor processing unit) is Google’s custom chip for AI work; the new systems give customers more specialized alternatives to NVIDIA’s general-purpose GPUs.

Federal regulators say Waymo robotaxis continued illegally passing stopped school buses after the company deployed a software update meant to fix the behavior. Filing says most reported incidents occurred after that update, putting the effectiveness of Waymo’s safety testing under fresh scrutiny.

NVIDIA’s board is investigating CEO Jensen Huang after he acknowledged an intimate relationship with Alice Huang, a Megaspeed executive whom U.S. officials suspect of working for Chinese intelligence. The case could complicate NVIDIA’s campaign to loosen restrictions on chip sales to China.

Introducing Box Architect, the next generation of enterprise AI. Box Architect enables teams to leverage frontier AI to create new workflows and apps, configure agents, connect to Box tools, and manage workflows at scale – all built on their trusted content and existing permissions in Box.

→ Design, Test, and Deploy with Box Architect

Commerce Department investigators spent months examining whether Singapore data-center company Megaspeed helped move advanced NVIDIA chips to China. The case was transferred to the Justice Department after NVIDIA’s general counsel contacted the agents and said a senior Commerce official had approved the outreach. NVIDIA says it routinely works with officials to answer questions and stop chip smuggling.

The unusual handoff raises a larger question about whether the government can police its most important AI supplier without giving that company too much influence over enforcement. The Justice Department is pursuing a criminal inquiry into Megaspeed. → Read the full New York Times report. (Paywall)

Ukraine is using AI to turn front-line data into possible battle plans and to scan reconnaissance-drone images for signs of enemy positions. One system builds a digital twin—a computer model of the battlefield—so commanders can compare possible actions before making a decision. A separate model looks for useful details in aerial imagery. → Read the full NBC News report.

Anthropic Buys $3.4B in Gas Turbines: Anthropic is buying Caterpillar generators capable of producing up to 1.7 gigawatts of electricity.

Meta’s New Lightweight Model: Muse Ember handles repetitive agent tasks like tool calls and document parsing, starting at $0.10 per million input tokens.

Claude Code Settings Expose Secrets on GitHub: Researchers found about 1,100 secrets in settings files 63 organizations posted to public repos.

SpaceX Lands $45B AI Compute Deal: Anthropic reportedly committed about $45B to use SpaceX’s Colossus data centers through May 2029.

Automattic Opens a Home for Vibe-Coded Apps: Spacefast offers databases, auth, file uploads, email and analytics for AI-built apps.

FAA Says AI Has Rerouted 180 Flights: SMART rerouted nearly 180 flights carrying passengers, cutting flight times 10+ minutes on average.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team