Copilot Fusion Picks Your Models, Agentic Workflows Hit GA, and VS Code Agents Keep Running After You Close the Window

At a glance

• Copilot Fusion now picks and combines models for each prompt, at three effort levels.
• Fusion bills every model it calls, so the cost of a task depends on its routing.
• Agentic Workflows turns a Markdown file into a GitHub Actions job run by a coding agent.
• VS Code agent sessions can now keep running in the background or in the cloud.

GitHub posted ten changelog entries between 11:30 AM and 12:00 PM ET Wednesday, and VS Code v1.144.0 followed at 2:51 PM ET, so nearly everything new today comes from one vendor. Our read: each change hands Copilot a choice you used to make by hand, which saves effort but moves cost and permission decisions into settings most teams have not opened yet.

Treat today as a Copilot settings review day: decide whether Fusion should become your team's default model, try one agentic workflow in a repo you can afford to experiment on, and update VS Code before you trust a long agent session to run in the background.

Top Stories

1. Copilot Fusion is GA and picks the models for each prompt

You select Copilot Fusion in the model picker, pick an effort level, and Copilot decides which models handle each prompt. Fusion Standard leans on versatile models for everyday coding, Fusion High on more capable models for harder reasoning, and Fusion Max on the most capable models for the most demanding work. Behind the picker it answers with one model, lets an efficient model draft and escalates to a stronger one if a quality check fails, or has a second model review the first before it revises.

GitHub says Fusion Max scored 60 on the Artificial Analysis Coding Agent Index, within two points of Claude Code with Opus 5.5 Max, at about half the average API cost per task and with about 40% less time per task. Those are GitHub's own figures, not independent results, so test them on your work. Billing is per model used, at each model's standard rate, with a limited-time 30% discount on model costs for paid plans. Fusion is available on all Copilot plans in VS Code, the Copilot app, Copilot CLI and JetBrains IDEs. Admins choose which models Fusion may use and can set Fusion as the default at any effort level through Enterprise managed settings.

A separate opt-in experimental preview lets Fusion route between on-device and cloud models. On supported Windows devices it can download MAI Code 1.1 Flash through Windows ML, and Copilot CLI on Windows, macOS and Linux can use MAI Code 1.1 Flash or Qwen3.8 Flash Next through a local OpenAI-compatible server. Routing decisions still happen in the cloud, so local inference does not make a session offline.

Practical dev impact: Because Fusion bills each model it calls, have one engineer run a week of normal work on Fusion Standard and compare their AI credit use with your current fixed model before an admin makes it the default for everyone.

2. GitHub Agentic Workflows is GA, so Markdown can drive agents in Actions

GitHub Agentic Workflows, an open source project from GitHub Next, lets you describe a task such as issue triage, CI failure analysis or docs upkeep in a Markdown file with YAML frontmatter. The gh aw CLI extension compiles it into standard Actions YAML, so it reuses your existing runner groups and policy constraints. The engine can be Copilot or another coding agent, and the quickstart lists Claude Code, Codex, Gemini and Pi.

Agents run with read-only permissions by default, inside a sandboxed container behind the Agent Workflow Firewall, and anything they propose goes through a safe outputs step and a dedicated threat detection job before it is applied. New since the public preview are self-hosted ARC runners with Docker-in-Docker, a gh aw forecast command that is no longer experimental and now covers threat detection credits and monthly ranges, model providers picked per workflow, a model router, and micro-VM isolation. The extension reached v1.0.0 on Monday and v1.0.1 on Tuesday, which added lock-safe pull request outputs and guided fixes for pull_request_target. It is available on GitHub Free, Team and Enterprise, and the default Copilot CLI engine needs a Copilot license.

Practical dev impact: Since workflows compile to ordinary Actions YAML, review the generated lock file in the pull request the way you review any workflow change, and run gh aw forecast before a scheduled workflow starts spending credits every day.

3. VS Code v1.144 keeps Copilot sessions running after you close the window

Cloud sessions run a Copilot task on GitHub's infrastructure, so closing your laptop does not stop it, and you can pick it up again in VS Code, the Copilot app, GitHub Mobile, github.com or Copilot CLI. Cloud sessions use a separate workspace, local uncommitted changes are not copied over, and the feature needs a paid Copilot plan. For local work, the new Agent Host keeps Copilot sessions going after you close a project folder or window, and Remote SSH and dev tunnels now reach the Agents window. Running code agent host --github-environment in a project folder lets you steer an agent on that machine from GitHub Mobile.

VS Code also published Agent Host Protocol v1.0, the protocol it uses to talk to its local and remote agents, with client libraries. Model comparisons are now on by default for eligible new Agents window sessions (sessions.chat.compareAgents.enabled) and computer use is experimental behind chat.agentHost.computerUse.enabled.

Practical dev impact: Background sessions now outlive the window, so check the sessions list before you start a duplicate task, and commit or push first if you want a cloud session to see your latest local changes.

Practical Impact Analysis

These three releases belong together because each one lets an agent keep working without you choosing a model, a runner or a window for it. Fusion leads because it changes billing for anyone who switches, and the only benchmark and cost numbers so far come from GitHub. Agentic Workflows deserves a test repo rather than a production rollout, because each scheduled run spends credits even with strong safeguards. Our read: VS Code v1.144 matters most to people who already run long Copilot sessions, since background and cloud sessions change what closing a window means.

Elsewhere at Universe, the Copilot app added Projects and computer use in public preview, plus a security extension for code scanning alerts installed with copilot plugin install github-security@awesome-copilot. Shared Copilot Automations are rolling out and bill AI credits to the repository owner rather than the person who created them, and GitHub Enterprise Cloud customers can get Copilot included through their account team.

Tutorial

1. Start the script from the root of a test repo where you have write access.
2. Step 1 prints versions, and step 2 reads the sample workflow the quickstart uses in a scratch folder.
3. Step 3 returns to your repo and looks for agentic workflow files.
4. Do the Fusion check in your editor, as described after the script.

# We have not run this script end to end on every platform. It only prints and checks unless you uncomment a line.
mkdir -p /tmp/universe-check && cd /tmp/universe-check

# 1. Which versions are out, and which do you have? At 5:52 AM ET Thu 29 Oct npm showed
#    @github/copilot latest v1.0.105.
npm view @github/copilot dist-tags.latest
if command -v gh >/dev/null; then gh --version; else echo "no gh CLI on PATH"; fi
if command -v code >/dev/null; then code --version; else echo "no code CLI, Agent Host needs VS Code v1.144"; fi

# 2. Read the sample workflow before you run anything. Look for the read-only permissions block.
curl -fsSL https://raw.githubusercontent.com/githubnext/agentics/main/workflows/repo-status.md -o repo-status.md
sed -n '1,25p' repo-status.md
grep -l "contents: read" repo-status.md

# 3. Back in your test repo. Lists any agentic workflow sources and their generated lock files.
cd - >/dev/null
ls .github/workflows/*.md .github/workflows/*.lock.yml 2>/dev/null
gh extension list
#    Optional gh extension install only. Pinned at 5:52 AM ET Thu 29 Oct, GitHub release channel:
# gh extension install github/gh-aw --pin v1.0.1
# gh aw add-wizard githubnext/agentics/repo-status
# gh aw forecast
echo "Done. Nothing was installed or changed."

For Fusion, open the model picker in VS Code or Copilot CLI, select Copilot Fusion, start on Standard, and note your AI credit use for a week.

Sources:

• Route and orchestrate models with GitHub Copilot Fusion
• GitHub Copilot Fusion now supports selection of local models
• GitHub Agentic Workflows generally available
• gh-aw v1.0.1 release
• gh-aw v1.0.0 release
• GitHub Agentic Workflows quickstart
• GitHub Copilot in VS Code, Universe releases
• VS Code v1.144 release notes
• VS Code 1.144.0 release
• GitHub Copilot app, Universe releases
• Security extension for the GitHub Copilot app
• Build shared workflows for your team with Copilot Automations
• GitHub Enterprise now includes access to GitHub Copilot

Recommended AI prompt

Copy this paragraph into ChatGPT, Claude, Gemini, Grok, or whatever AI you use.