Walmart puts a leash on your shopping agent.

Microsoft's diagnostic AI just achieved 83% accuracy on complex medical cases in a simulated environment. For context, the physicians in the same simulation managed a dismal 36% accuracy, even with access to reference materials.

This isn't a marginal victory; it is a blowout. A simulation is not a hospital ward, and synthetic benchmarks rarely survive contact with messy real-world patient data. But a 47-point gap means the models are seeing patterns human specialists are missing entirely.

Physicians lose their monopoly on complex diagnostics the moment these systems get clinical approval. The immediate winner is Microsoft, which now has the ultimate wedge to sell enterprise healthcare systems. The losers are the regulatory bodies trying to justify why a system that doubles diagnostic accuracy shouldn't be the default second opinion.

Legora just closed a $550 million funding round, vaulting the legal AI company to an $8.8 billion valuation. The capital is earmarked for expanding their legal AI business across the U.S. and global markets.

You don't get an $8.8 billion valuation by building a slightly better search engine for case law. Legora's backers are betting that AI can fundamentally replace the grunt work currently billed at $400 an hour by first-year associates.

Law firms that adopt these tools will see their margins explode, while firms clinging to manual review will get priced out of the market. Legora wins big here, but the real test is whether their models can handle the paranoid precision required by global compliance regimes without hallucinating a precedent.

Flapping Airplanes just raised $1 billion to develop AI systems that can learn using significantly less data and computing power. The goal is to build more efficient and accessible AI models that don't require draining a small lake to cool their data centers.

The industry is obsessed with scraping the bottom of the internet to feed massive clusters, but Flapping Airplanes is betting that algorithmic efficiency will beat sheer scale.

If they pull this off, the moat for companies like OpenAI and Google shrinks dramatically. Models that require less compute lower the barrier to entry for everyone else. The incumbents lose their structural advantage if intelligence no longer requires a billion-dollar supercomputer to train.

GM tested over 300,000 battery designs entirely through computer modeling before they even bothered to construct a physical prototype. The AI-powered simulation drastically accelerated their development timeline.

Physical prototyping in the automotive sector is brutally slow and expensive. By shifting the trial-and-error phase entirely into a simulated environment, GM can iterate at the speed of software rather than the speed of manufacturing.

The winners are the engineering teams who can now validate material sciences overnight. The losers are the traditional automotive suppliers who still rely on physical bench testing to validate their components. Digital prototyping is no longer a luxury; it is the baseline for survival.

Loom just launched a 'Record for agent' feature in open beta for Business + AI and Enterprise users. The tool converts narrated screen walkthroughs into explicit AI agent instructions, complete with screenshots and links.

Writing prompt sequences for GUI agents is tedious and brittle. Loom's approach lets operators do what they already do—record a quick video—and lets the software extract the precise DOM elements and execution steps required to automate the task.

This removes a massive bottleneck for builders trying to deploy agents in non-technical environments. Loom wins by embedding itself as the translation layer between human intent and agent execution.