AI roundupMonday 9 November 2026

Meta pauses AI training terms after backlash

Meta's attempt to rewrite its terms of service to scrape the web for AI training ran into a wall of user anger. The company paused the change, but the episode shows how quickly trust erodes when companies try to claim broad data rights. Meanwhile, the practical work of building and deploying AI agents continues in dealerships, bookkeeping, and code refactoring.

In today's edition: Data rights and pushback, Agents in the wild and Security and governance, plus quick bites.

Data rights and pushback

The day's biggest story is about who gets to use your data for AI training, and what happens when they try to take it.

Agents in the wild

While the policy fights play out, companies are quietly putting AI agents to work in specific, measurable roles.

startuphub.ai · 2 min

0
Ramp built evals for months so AI can close books

Ramp spent four to six months building evaluation metrics before launching its AI bookkeeping agent, Stacks. The company created synthetic businesses to test accuracy without using live customer data, and found that reasoning capability was the hardest challenge for the AI.

simonwillison.net · 1 min

0
Comment: The Web Is More Than Enough, by Paul Ford

Simon Willison shows how Claude ported a Three.js project to Vanilla JS with almost no changes. It's a small demo, but it illustrates how LLMs are becoming reliable tools for code refactoring within a single chat thread.

Security and governance

Two stories today underline that the infrastructure around AI, code, data, board oversight, is still catching up to the pace of deployment.

Quick bites