May Product Update

Aaron Smallberg -

Here's everything Plural shipped in May.

🔧 Workbench Agent Improvements

We released Workbenches last month, for Plural's Build-Your-Own Agents for DevOps. In May, we added a ton of new features and improvements to Workbenches:

  • Workbench Evals. Every Workbench run can now be graded end-to-end. The eval panel produces an overall score alongside a written summary, a breakdown of the prompt's clarity and scope, a critique of the conclusion against the evidence the agent collected, and a "logic and thoughts" trace that walks the agent's reasoning step by step. Failed steps and missing investigation paths are called out explicitly, and the panel exposes a "Create skills from eval" button so a recurring weakness becomes a new skill the next run inherits.
  • Babysitting mode for agent runs. A new babysit flag on agent runs keeps the agent active after it opens a PR so it can monitor review feedback and requested changes, then follow up until the PR is ready to merge. Surfaced as a checkbox in the agent-run UI for Write-mode runs.
  • Agent MCP server as a sidecar. The agent-harness MCP server now runs as a sidecar with its own bootstrap binary, isolating tool execution from the agent's reasoning loop and making the harness pod easier to operate and debug. Includes a new dockerfiles/mcpserver/agent image and docker-compose for local development.
  • Git-based skill updates. Workbench skills can now be pulled and updated from a git source, so a team's runbooks and remediation patterns live alongside the rest of their GitOps state instead of in a separate config surface.
  • Improved integrations: Full GitLab, Bitbucket, and Azure DevOps SCM support. CloudQuery is now in production.

🔔 Sentinel Improvements

Postrun scripts. SentinelCheckIntegrationTestConfiguration now supports postrun scripts — arbitrary commands that run after a sentinel check completes, useful for posting results to external systems, cleaning up test fixtures, or triggering downstream work without standing up a separate webhook.

🏢 Company Updates

We’ll be attending KCD NYC and AWS Summits in NYC & DC in June. Let us know if you plan on attending any of those events!