Governance gets talked about a lot. Sometimes it's a committee. Sometimes it's a document that lives in a shared drive nobody opens. But if you're an engineer or an architect, governance needs to mean something more concrete than that. It needs to show up in your runtime, in your code, in the boundaries you actually enforce.
That's what I want to walk through here.
From answers to actions
We've made real progress giving AI the ability to reason. It can retrieve information, synthesize context, and now, with agentic AI, it can take action too. That last part is where things get genuinely interesting, and genuinely risky.
Think about what's changed over the last couple of years. The same models that used to generate text can now select tools, initiate workflows, update CRM records, send messages, and trigger transactions. We stopped building systems that give answers. We started building systems that take actions on our behalf.
That shift, from answering to acting, is the context for everything I'm going to cover.
Meet Alex
Let me introduce a fictional agent called Alex. If you have a real agent at your organization, keep it in mind as you read, because the story will probably feel familiar.
Alex started with a sensible job. It found customer information, summarized interaction history, and helped employees respond faster. The employee still made the final call. Alex just prepared the information.
Alex worked well, so the engineering team decided to connect it to internal API services. Now Alex could complete tasks automatically, without a human copying data from one screen to another. Reasonable enough.
Then a customer asked about their insurance coverage. Alex found the correct record, identified the right tool, and made a change to the coverage record. The API returned a 200 OK. Everything looked fine. Alex told the customer their request had been completed.
Here's the problem. The customer had asked a question. They never asked for a change to be made.
No one was hacked. The credentials were valid. The endpoint behaved exactly as it was designed to. The logs captured everything. And yet, the agent made a change that no one in the business had authorized it to make.
That distinction is where governance has to start.
For expert advice like this straight to your inbox every other Friday, sign up for Pro+ membership.
You'll also get access to 300+ hours of exclusive video content, a complimentary Summit ticket, and so much more.
So, what are you waiting for?
Capability and authority are different things
Here's a framing I keep coming back to. The boundary between what an agent can do and what it should do is not enforced by the model. It has to be enforced by the system around it.