Copilot delegates model choice to Fusion and expands enterprise access

GitHub made automatic model routing generally available through Copilot Fusion and included Copilot access for eligible GitHub Enterprise customers. Coding agents also moved further into repository maintenance and security remediation, with new releases from GitHub and Cognition.

Key developments

Fusion takes over model selection

GitHub Copilot Fusion became generally available. Users can select Fusion instead of choosing a model for each task, letting Copilot route and orchestrate models automatically.

Developers can delegate model selection as part of the coding task rather than interrupting work to choose a model.

Oct 28–29 · Use Case: copilot - GitHub Changelog

Copilot joins eligible Enterprise plans

GitHub made Copilot available as part of GitHub Enterprise for every licensed user at enterprises working with a GitHub or Microsoft account team.

Eligible enterprises can extend coding assistance across their licensed users without treating Copilot as a separate purchase.

Oct 28–29 · Use Case: copilot - GitHub Changelog

Devin connects security scans to fixes

Cognition announced Devin Security 2.0, combining frontier cyber-model scans, environment-aware prioritization and autonomous remediation that turns findings into pull requests. It also announced hundreds of millions of dollars in Devin credits to help enterprises find and fix critical vulnerabilities.

Security findings can move directly into proposed code changes, reducing the manual handoff between discovery and remediation.

Oct 26 · Devin Blog · Blog | Cognition

Repository workflows reach general availability

GitHub Agentic Workflows became generally available. The open-source GitHub Next project automates reasoning-based tasks including issue triage, CI failure analysis and documentation updates.

Recurring repository maintenance can be delegated to agents alongside code-writing tasks.

Oct 28–29 · Use Case: copilot - GitHub Changelog

Copilot permission assistance expands

Assisted permissions became generally available in GitHub Copilot CLI, the Copilot app and Visual Studio Code. The feature applies when Copilot requests permission for actions such as running commands or accessing file-system paths.

Developers using all three interfaces gain permission assistance at the point where agents request access to commands or files.

Oct 26 · Use Case: copilot - GitHub Changelog

Agent Skills moves toward neutral governance

Anthropic announced it is donating the Agent Skills standard to the Agentic AI Foundation under the Linux Foundation, with a commitment to open, community-driven and vendor-neutral governance.

Reusable agent skills gain a governance path outside a single model vendor.

Oct 29 · Simon Willison

Claude Code broadens hosted monitoring

Claude Code v2.1.310 added the Monitor tool and plugin monitors on Bedrock, Vertex and Foundry, including configurations where telemetry or nonessential traffic is disabled. Long foreground sleeps are now blocked in those environments.

Enterprise users with restricted telemetry can use monitoring features across more hosted deployment configurations.

Oct 29 · Release notes from claude-code

Codex adds GPT-6.1 Terra

Codex 0.166.0 added GPT-6.1 Terra to its bundled, Vertex AI and Amazon Bedrock model catalogs. It also introduced keyboard-driven transcript selection with integrated search and configurable shortcuts, plus more consistent Markdown rendering.

Codex users gain another model option and easier navigation through long coding-session transcripts.

Oct 30 · Codex releases

Maintainers judge agent output

Cognition introduced FrontierCode 2.0 with more diverse, challenging tasks from more repositories. Open-source maintainers grade whether agents’ output is actually mergeable.

The benchmark tests whether generated changes are acceptable to maintainers, not just whether they pass automated checks.

Oct 27 · Blog | Cognition

Also this week

Engineering practice

• OpenAI described its engineers using Codex Ultrafast to investigate incidents in minutes rather than hours, alongside security analysis, refactoring and daily engineering work. OpenAI News

• Greptile is using GPT-6 Astra to test user workflows, Daybreak to strengthen security and Codex to reduce rework, according to an OpenAI account. OpenAI News

• CodeRabbit uses GPT-6.1 Sol to investigate security risks across complex codebases, with developers reviewing its security agent’s findings and proposed fixes. OpenAI News

• An offline coding-agent walkthrough demonstrated a setup using OpenCode, Ollama and Qwen. Latent Space

Enterprise and infrastructure

• OpenAI and ServiceNow announced an expanded partnership bringing OpenAI models and agents into ServiceNow, and ServiceNow workflows into ChatGPT and Codex. OpenAI News

• Anthropic engineers Caitlin and Angela described managed agents that separate a durable harness from sandboxed execution and keep credentials outside the agent environment, while also discussing the completed Rust rewrite of the Messages API and long-running, multi-agent workloads. Latent Space · Latent Space

• OpenAI launched an API-credit benefit for eligible ChatGPT subscriptions, following a similar Anthropic offering. Simon Willison

Agent-built games

• Simon Willison argued that October marks an inflection point for agent-assisted game development, citing Tree Appreciator, Emberwick and Battle Drones, and added a four-player online map to Battledrones, whose previous maps supported two players. Simon Willison · Simon Willison

• Jacob Sansbury described building Spawn, a massively multiplayer environment where players prompt an in-game agent called Savi to modify the shared world. Simon Willison

What to watch

• Watch for completion of the Agent Skills donation and its community-governance arrangements; timing for this and the following announced steps was not given.

• Watch for the rollout of ServiceNow workflows in ChatGPT and Codex under the expanded partnership.

• Watch for access to Cognition’s announced Devin credits for enterprise vulnerability discovery and remediation.