Glean adds headless agent building for Claude Code, Cursor, and Codex

The agents are defined by specific files, including a spec.yaml for configuration and an instructions.md for behavior, which can be stored alongside other code. Despite being built in external coding assistants, these agents run on Glean’s managed runtime, retaining access to the company’s knowledge graph, permissioned access to over 275 enterprise sources, and full audit trails. Glean states this approach provides deterministic control over agent steps and guardrails, appealing to teams that require rigorous software engineering practices for AI deployment.

The feature supports Auto mode agents, while Workflow mode agents must still be created in Glean’s visual builder. Developers install a specific Glean plugin for their chosen environment, authenticate via OAuth, and use a dedicated command to run and publish agents. Glean positions this as a way to apply familiar testing and review processes to AI agents, ensuring security and governance remain intact while offering the flexibility of a code-first workflow.