How are people thinking about governance as custom agents scale?

Smart CRM & Sales Hub Breeze & Other AI-Powered Features
,
1

I have been testing the new Agent Hub and Agent Builder beta, and one question keeps coming up.

Building the first few agents is relatively straightforward. But what happens later when a portal has many agents built by different teams, each with different tools, write access, guardrails and assumptions about the business?

A few things I am curious about:

  • Who should own an agent after it goes live?
  • How often should tools and write permissions be reviewed?
  • How do you avoid conflicts between agents, workflows and integrations?
  • How do you decide what should be an agent versus a deterministic workflow?
  • What should an agent do when it reaches the edge of what it can safely complete?

We put together a small prototype while exploring this. It looks at agent inventory, ungated write actions, approval controls and escalation when an agent should not continue on its own.

I would genuinely be interested to hear how admins, partners and RevOps teams are approaching this.

2

This is where things can get messy fast. I’d start by being very careful with write access. Any agent that can change records should have a clear owner who is responsible for it. For me, if the logic can be written as a fixed rule, it should probably stay a workflow. It is much easier to understand and audit later. And when the agent is not sure, it should stop and ask instead of guessing. A short delay is better than the wrong data being written into the CRM.