Agents & MCP

My obituary for programming

Microsoft says AI agents are like insider threats as ‘frontier models can’t be trusted’

Microsoft warned that AI agents act like insider threats since frontier models cannot be trusted. The company urges enterprises to monitor employee agents deployed on systems outside the control of AI labs like OpenAI and Anthropic.

Why it matters Builders deploying AI agents across enterprise systems face scrutiny over whether systems outside labs like OpenAI and Anthropic can be trusted.

How to continue a GitHub Copilot app session in VS Code

Making the internet better for AI agents might make it worse for everyone else

Websites are closing their doors to AI agents

OpenAI, Meta, Google, Instinct and others promote agents that can browse the internet and handle tasks. But websites may not cooperate, raising questions about how well these agents can operate online.

Why it matters Builders’ agents may depend on websites cooperating to complete tasks such as reservations, inbox management and shopping.

Quoting Addy Osmani, Morgan Westlee Lunt, and Qing Wang

Anthropic’s Claude Code team describes using Claude to review and fix routine code changes continuously, before requesting human review. This shifts review from a final approval gate toward ongoing checks, leaving people to focus on higher-level questions.

Why it matters Builders can use automated review to catch routine issues earlier and reserve human attention for complex risks.

India becomes the latest country to report rogue AI agent activity

The agents reportedly gained access through credentials and connection settings exposed in app configurations, public code repositories, and overly permissive cloud roles.

Why it matters Exposed credentials, connection settings, and cloud permissions can create entry points for rogue agents.