Copilot's six updates bundled and launched; the most practical one is the one about halving costs

A week after GitHub Universe ended, GitHub bundled and shipped six Copilot updates. Several are about 'keeping agents working continuously,' but the first thing worth looking at is cost.

Copilot Fusion: automatic model selection, lower cost

Fusion is a runtime model router and orchestrator. You give it a task; it chooses the model and execution mode, then returns a result. Citing data from Artificial Analysis's Coding Agent Index, GitHub says Copilot Fusion Max scores within two points of Claude Code paired with Opus 5.5 Max, but the API cost per task is roughly half.

From now until November 30, Fusion is 30% off.

Copilot Projects: bring parallel work into one place

When multiple repos and multiple agents are running at the same time, goals, context, sessions, and results easily get scattered. Projects gathers these together, making it easier to track progress and decide the next step. GitHub's Kyle Daigle said he tried things out and ultimately found it smoothest to dump his ideas to the agent all at once in Projects; Burke Holland put it more directly: an agent that manages agents. It works with both local and cloud agents.

Security extension: code and alerts on the same canvas

GitHub added a security extension to the Copilot app. Code you and the agent just wrote is placed in the same Security canvas as the repo's code scanning alerts. You can have Copilot explain and fix the issues it finds, then review the changes one by one.

Copilot Dreaming: let agents remember the past

Dreaming reviews past work to improve later agentic sessions. It can add, update, and delete personal memories and repository memories. It is currently in public preview. Memory has always been tricky for agents: too much is noise, too little is as good as none, and the real effect will have to be judged after using it for a while.

Copilot Automations: run agent tasks on a schedule or on demand

You can create automations for repetitive agent tasks, run them on a schedule, or trigger them manually. Teams can also share automations: visible to repo readers, editable by writers, and AI quota is charged to the repo owner.

Cloud tasks: start anywhere and keep going

Cloud tasks can be started from the GitHub website, GitHub Mobile, Copilot app, Copilot CLI, @code, or conversations in Slack and Microsoft Teams. You can start on any client and continue somewhere else.

None of these features is revolutionary on its own, but together they point in the same direction: turning agents from 'single conversations' into 'continuous projects.' Fusion reduces cost, Projects manages context, Dreaming manages memory, and Automations manages repetitive tasks. For people who work with agents every day, the ones worth trying first are Fusion and Projects: one saves money, the other saves hassle.

The original article and the full list of updates are in GitHub's October Ship Log: https://x.com/i/article/2117339128920752128