Mistral turns Vibe into a work agent

Mistral AI has presented Vibe as the new form of Le Chat, with a clear product shift: one agent for long-running office work and software development. The official post describes a service that is already live, carrying over Le Chat conversations, settings and plans while adding two more structured surfaces: Work Mode for multi-step business tasks and Code Mode for programming sessions launched from the web, an editor or the terminal. In practice, the company is positioning Vibe less as a chat box and more as a controlled workspace for delegated tasks.

The notable point is not only the rebrand. Mistral is bringing together three uses that many vendors still split apart: searching across company knowledge and messages, producing deliverables such as reports or dashboards, and changing code all the way to a pull request. In Work Mode, Vibe can connect to Google Workspace, Outlook, SharePoint, Slack, GitHub and custom connectors. It can analyze a database or spreadsheet, draft a document in a canvas, and schedule recurring prompts on a daily, weekly or monthly cadence. That scheduling feature matters because it moves the agent from one-off answers toward repeatable business processes.

For developers, Mistral is adding a dedicated web surface for code, a VS Code extension and updates to the Vibe CLI. Code Mode starts isolated sessions where users can inspect diffs and keep control over sensitive actions. The extension is designed to read, edit and run commands inside the open project, with context pulled from files as well as GitHub, GitLab, Jira or Linear. Sessions can persist when the user’s machine is offline and can run in parallel, putting Vibe closer to the category of asynchronous development agents than to a simple completion assistant. The CLI updates also add skills, custom modes, subagents, editable plans and session-scoped permissions.

The broader issue is the move from a general chatbot to a work system governed by permissions, connectors and execution traces. Mistral emphasizes visible steps and expandable tool calls, which matters for organizations that want automation without giving up auditability. The launch does not prove that the agent will handle every complex enterprise workflow reliably. It does show where competition in AI agents is heading: toward practical orchestration, where documents, code, permissions and scheduled tasks are connected in one controllable flow rather than scattered across separate assistants.