AI Round-up: 18 October 2026

Website access and delegated authority are the main agent stories: The Verge reports that sites are blocking AI agents, while the Financial Times examines Anthropic's work with lawyers on machine-led negotiation. Coverage of Anthropic's own workforce shows another side of adoption: a smaller share of junior technical roles and more human attention devoted to coordinating and reviewing agent output. These company findings are not evidence of an economy-wide employment effect, and Jack Clark's growth projections remain forecasts.

GitHub highlights Logitech keypad controls for Copilot workflows in the snapshot's sole qualifying X post. YouTube coverage combines agent-safety and funding commentary with practical Claude website-building, ChatGPT interface and assistant workflows. Reddit's retained discussions focus on hardware for Perplexity Computer and building agents and flows in Copilot Studio.