The News, Explained
GitHub published its weekly Copilot release update on October 9, 2026. Rather than centering on a single new model, the release changes how developers control an AI agent’s local access and manage multiple accounts, models, and work sessions. Claude Haiku 5.5 is now available to Copilot Pro, Pro+, Max, Business, and Enterprise users. Source
The largest operational change is the general availability of local sandboxing. A sandbox is a bounded execution area that limits which files, networks, and credentials an agent can reach instead of giving it unrestricted access to the computer. It is available in Copilot CLI, the Copilot app, and VS Code sessions that use Agent Host, and it is included with Copilot at no extra cost. The announcement describes controls that limit access; it does not claim that sandboxing removes every risk. Repository secrets and network permissions still need to match the actual task. Source
The Copilot app can now use different GitHub accounts for the Copilot license and repository access. For example, a developer can authenticate a company-provided Copilot license with one account while working with repositories available to another account. This avoids requiring the license and every repository permission to sit on one identity. The repositories and actions available still depend on the permissions of the selected repository account. Source
In Copilot CLI, the /model command can discover supported models from a running local Ollama instance. Ollama is a tool for running AI models on a user’s computer or a server they manage. The discovered local choices appear alongside configured models and cloud models supplied through GitHub Copilot, so a developer can inspect model options without leaving the terminal. The announcement does not establish the installation, performance, or hardware requirements of each local model. Source
VS Code 1.141 can arrange multiple agent sessions side by side in a grid, making it easier to compare outputs or follow several tasks. The Open worktree cleanup action shows how much disk space inactive session worktrees use and lets the developer choose which ones to remove. A worktree is a Git feature that checks out another branch into a separate folder, allowing parallel work on one repository. More parallel sessions can therefore leave more worktree folders on disk. Source
OYOPICK’s Take
The common thread is operational control rather than model quality alone. Sandboxing, account separation, local-model selection, and worktree cleanup all help answer the same question: which agent is doing which job with which permissions and resources? As agents move across repositories and tools, those controls may become as important as switching to a more capable model.
More choice also creates more configuration responsibility. A split between a company license and another repository account should comply with the organization’s access policy. A local Ollama model still needs its quality and speed tested on the same representative task. A sandbox that is too broad weakens its boundary, while one that is too narrow can block necessary work. Starting with a representative repository and limited permissions gives a team a practical way to adjust file, network, and credential access.
For AI development tools to remove repetitive work for more people, higher capability needs to arrive with controls that users can understand and reverse. We hope this release leads beyond running more agents at once toward a clearer record of what each session accessed and which local resources it left behind.