AI in insurance and pricing, the Liquid AI model, and tools for agents

In today’s roundup: insurers’ statements about additional costs due to AI and McDonald’s plans to assess customers’ willingness to pay. Also — a new Liquid AI model, an analysis of AI effectiveness evaluation, and community finds.

🔥 Hot:

🔹 Blue Cross insurers said AI tools caused nearly $1 billion in additional costs — Reuters reports their estimate: AI adoption resulted in new expenses for the companies. 🔹 McDonald’s wants to use AI to assess customers’ willingness to pay at each restaurant — The idea is to account for differences between customers and locations.

➡️ News:

🔹 Liquid AI introduced the decision model D1 🔹 Several platforms in China are competing to become the country’s own “Hugging Face” — The analysis focuses on the ModelScope and MoArk ecosystems.

➡️ Useful materials:

🔹 An analysis explains the AI productivity paradox in companies — Processes seem faster internally, but customers do not always notice improvements — the author suggests considering how to evaluate the impact of AI initiatives.

➡️ Discussions and case studies:

🔹 A user ran Qwen3.8 Flash Next on a laptop with 12 GB of VRAM via Strata — The author reports a generation speed of 50 tokens/s; the engine currently works on Nvidia, while AMD support is experimental. 🔹 A developer built WinMind — an MCP server for controlling Windows through the accessibility tree — Instead of screenshots and coordinate-based clicks, the agent accesses interface elements through Windows UI Automation. 🔹 Runtape proposes debugging AI agents with counterfactual scenarios and regression tests

📝 If you would like to add other news and materials to the list, write in the comments.