Glean Agents adapt to real-world complexity and are built to scale safely across your enterprise

Every organization is inherently unique—defined by individuals, data, and processes that reflect company goals, competitive dynamics, and market influences. Since these elements are constantly changing, agents cannot remain static; they must adapt dynamically to real-world complexity. As they evolve, robust guardrails are essential to ensure agents access only the right data, interact with the right people, and remain aligned with organizational governance. When implemented with proper control and oversight, agents can transform how businesses operate. 

Today, we’re rolling out new capabilities in Glean Agents that make it easier to build and safely scale agents that adapt to real‑world complexity. Our unified agent builder now combines graph-based and conversational builders to deliver a more powerful creation experience. New Fast and Thinking modes in the plan & execution node give agents the ability to solve more complex problems, while users can optimize for speed versus depth in the response. More granular agent permissions and routing capabilities offer admins the tools to better preserve trust, enforce permissions, and prevent sprawl.

A unified agent builder 

After introducing Glean's conversational builder at Glean:LIVE Fall’25, we heard from experienced agent creators that they wanted access to its capabilities, even when creating agents in Glean’s agent builder. Today, we are launching a unified interface that brings our graph-based and conversational builders together. The agent builder now lets you move between natural language agent creation and directly building your agents step by step. You can chat to refactor the entire agent graph—add or remove nodes, edit instructions, learn about individual steps, and even debug. Start in natural language, take manual control where it matters, and scale from simplicity to power-user depth on demand.

Plan & execute steps now include Fast and Thinking modes 

We’re upgrading the plan & execute step in Glean’s agent builder to become an increasingly agentic node that takes agency to do the job it was tasked to complete. In the plan & execute steps, agent builders can choose between Fast or Thinking to control how much “effort” an agent puts towards the task—enabling builders to optimize for speed versus depth while leveraging the upgraded reasoning capabilities of Glean’s Agentic Engine 2. This already highly autonomous node now has increased reasoning power and the ability to adapt to the world around it. In both modes, the agent proposes a plan, executes across the enterprise tools you enable, asks for help when clarification is needed, and adaptively plans and re-evaluates as it uncovers new context. This makes it capable of delivering more complete, higher‑quality answers even in ambiguous situations.

With these updates, agents can now handle more complex tasks that require multiple steps or decisions while continuously anchoring to the objective you want to accomplish—like analyzing data or building comprehensive reports that need information across various sources within a single step. In internal testing, both modes improved answer completeness with Thinking mode reaching 94% end‑to‑end answer completeness for our most difficult production use cases. 

More granular agent permissions for enterprise-grade control

While the ability to build an agent should be easily accessible for everyone, agent sprawl should be controlled. With more granular agent sharing permissions, admins can control who Agent Creators can share agents with by defining Default Member permissions that limit access to agents based on specific individuals, departments, or an entire company. These Default Member permissions also determine whether each group can publish agents. 

We’ve heard from customers that they need this increased granularity as they scale agent rollouts across their company. Not only do these increased permissions help with rollouts, they empower creators to be able to build specialized agents, test them, and share their production-ready agents with targeted groups of relevant individuals. Sensitive and non-relevant agents stay restricted, while high‑quality, broadly applicable agents are easy to find and reuse.

Agent routing from Assistant

Employees shouldn’t have to hunt for the right agent. With agent routing from Assistant, admins configure rules that automatically send certain questions to specialized agents. Users keep a single entry point—Glean Assistant—and get authoritative answers behind the scenes from the right expert: HR policy questions to your HR agent, access requests to your helpdesk agent, product questions to your roadmap agent, and more. Employees get faster, higher‑quality answers without changing their habits.

Agents that adapt, scale, and improve with you

These capabilities accelerate how agents are securely built and scaled to drive real impact. For agent builders, a unified agent builder speeds up creation, while Thinking mode in the plan & execute step produces sturdier agents that can handle ambiguity. For IT and admins, more granular agent permissions and agent routing from Assistant deliver governance and discoverability at once—so you can scale usage safely without clogging the library or risking oversharing.

Learn more about how to get started with Glean’s agentic features through these resources:

• Getting started with Agent Builder
• How to use Plan and Execute steps
• Managing agent routing capabilities

Check out other details on exciting new Glean features coming your way this week, or sign up for a demo to learn more about Glean!