Experience so far

amilazzo

• I have been using Claude Code heavily with the Glean MCP, I've been looking for a way to replicate that power-user experience natively in Glean. Autonomous Agents help bridge that gap by providing an unbounded agentic loop. So far, I've tested Autonomus Agents with a use case that performs many different searches for Jira ticket hygiene.

• What I like:

• - It is able to gather and process a larger, more exhaustive list than a workflow agent, such as dozens of tickets

• - Refining prompts feels more natural than refining the workflow. Using natural language doesn't just guide output style, but actually instruct it to behave differently (ie calling tools)

• - The complexity of maintaining the agent doesn't go up as I make it do more complex things

• What I don't like:

• - Still very limited by the action count and what Glean deems is "auto-runnable" - I would love to be able to close Jira tickets out, but that's not supported, and Open tickets, which is not available for autonomous usage, edit documents rather than create new ones each time

• - It would be great to have a canvas or some equivalent to iteratively add data to as it works through the problem you give it, I typically use a Markdown file for this when working it with Claude Code

• Again, I'm comparing it to Claude Code which may not be entirely fair. That said, it would be nice to have that level of "power user" experience in Glean where it can easily whitelist "safer" actions, edit files, combined with the governance and ease of use of Glean's managed web experience. Compared to the existing Glean constructs such as workflows, this is an improvement for many use cases.