AI’s next test: trust, teaching, and choosing sides

A discussion featuring AI for Education CEO Amanda Bickerstaff and Philadelphia teacher Kate Conroy points to a more practical phase of classroom AI: focusing on its benefits while grappling with how teachers and students should use it responsibly. For builders, the challenge is creating tools that support learning without quietly replacing the thinking educators are trying to develop.

The U.S. is reportedly preparing to tell dozens of countries that they must choose which side of the AI competition with China they will align with, including possible consequences for those joining a rival framework. Anyone building or deploying AI internationally should be watching this closely: access to chips, models, partnerships, and infrastructure may increasingly depend on national alignment.

Google’s Gemini 3.7 Flash is positioned around coding, software engineering, and AI agents, with improved performance at lower cost. That combination could matter more than benchmark headlines for developers: cheaper inference can make experimentation and agent workflows viable for smaller teams, provided reliability keeps pace.

Columbia Law School has introduced a largely permissive policy allowing students to use generative AI as a learning aid. The important question for educators everywhere is not simply whether AI is allowed, but how students can demonstrate genuine understanding when drafting, feedback, and explanation are increasingly shared with software.

A new analysis argues that AI’s bottleneck is shifting from producing answers to checking whether those answers are correct, with India highlighted as a place where adoption is outpacing verification capacity. For tool users, this is a useful reminder that a fluent response is only a starting point—and for builders, verification needs to be designed into the workflow rather than added as an afterthought.