Simon Willison highlights Jacob Sansbury's account of building Spawn, a multiplayer game world changed by prompts to an in-game AI. The team says a swarm of Fable-powered agents edits live code and ships to production about hourly. It is a rare, concrete look at how far coding agents can run autonomously in production.
Amp can now boot a cloud Mac during a thread to build, test and run iOS and macOS apps. The feature, macOS Satellites, is in beta and billed hourly. For developers working on Apple platforms, it changes how an agent verifies its own changes.
The GitHub Copilot app now offers Projects in public preview, a home for work that spans several repositories, sessions and non-code assets. Copilot can pull in Slack and Figma context and run multiple agents that return pull requests for review. A computer use preview on macOS and Windows lets it operate desktop apps that lack an API or MCP integration.
GitHub introduced GitHub intelligence, a context layer for Copilot agents. Copilot Memory is now generally available on paid plans, dreaming is in public preview, and Copilot can search code across repositories without a local clone. Agents can now keep your preferences and team conventions between sessions and reach related projects.
GitHub is rolling out shared Copilot Automations, so a repository's cloud agent workflows no longer depend on the person who created them. Teams can read, edit and keep running these automations, with AI credits billed to the repository owner. Useful for bug reproduction, PR screenshots and incident triage.
GitHub has expanded how Copilot agents run in VS Code. Cloud sessions can be picked up across VS Code, GitHub Mobile and github.com, while the new Agent Host keeps local and remote agents running after you close a window. Computer use and side-by-side model comparison arrive as preview or experimental features.
GitHub Copilot now supports server-to-server authentication for GitHub Apps installed on organizations, in public preview. Agents and services built with the Copilot SDK can make requests for an organization without a signed-in user's credentials. Usage is billed to the owning organization and governed by its model policies and budgets.
The article examines what a .NET AI agent does when its own guardrails fail, such as a throwing guard, a down classifier, a missing caller or miswired middleware. The author deliberately breaks each guard layer and shows how to make the request blocked instead of silently allowed. It ends with a practical checklist for developers building guarded agents.
Amp now lets you run its orbs, the remote machines where coding agents do their work, on your own infrastructure. The amp orb provider command relays spawn, pause, and delete requests from Amp to a dispatch program you write. Teams that want agents on their own servers should note the setup model and the October 29 pricing change.