Glean · Week of Nov 9–15, 2026
This week so far
Onyx Security integrates its agent-governance control plane with Glean agents
Onyx Security announced on Nov 13 an integration that lets enterprises inventory and govern the AI agents they build and run on Glean. Onyx can discover each Glean agent and its access, assign an owner, business purpose and risk class, and track changes to its permissions, tools and data access over time. Runtime controls on prompts, responses and actions can block unsafe requests or steer agents back to policy, with central audit evidence kept across environments. Glean's partnerships VP framed governance as foundational to trustworthy enterprise AI; the news was shared on X and Bluesky with little discussion.
Glean brings its assistant into Microsoft Excel as an add-in
Glean announced Glean in Excel, an add-in that opens a task pane next to a workbook so users can ask in plain language for formulas, calculations and charts while pulling in company knowledge they can already access in Glean. Glean's docs describe it as a beta deployed by Microsoft 365 admins through the add-in portal, working in Excel on Windows, Mac and the web. The add-in requests read-write access to the open workbook and can edit cells, add sheets and tables, and read more of the file when it needs context. The docs also flag that SharePoint site allowlists and denylists are not yet fully applied inside the add-in. Reaction on X was light, mainly the company's own launch post.
Glean explains the eval system behind its automatic model routing
Glean published an engineering and product write-up on how its auto routing picks a model and reasoning level for each Assistant request. The company says about a quarter of requests are simple lookups sent down a fast lightweight path, while harder work such as dashboards, exhaustive analysis or dissatisfied follow-ups is handed to task-specific expert models along with the context already gathered. Its routing choices rest on replaying 5,000 real enterprise tasks across 37 model and reasoning configurations from 11 models, judged blind by an agentic grader; in one sales task set a frontier model cost five times more with no quality gain. Admins pick Economy, Balanced or Frontier policies per department, and Glean's post promoting it drew modest, positive replies on X.
No Director, VP, or CTO postings this week.