It's easy to build a platform that controls people. It's harder to build one that makes doing the right thing the easy thing.

After many years of cloud infrastructure and platform work, I'm convinced the biggest obstacle to good platform operations isn't technology, it's friction. When it's easier to bypass the platform than to use it, people do. And then it hardly matters how good your guardrails are.

The right thing has to be the easy choice. If it isn't, the platform has already lost before it started.

What I work toward is self-service with built-in guardrails. Not a team sitting and approving pull requests, but systems that make the right thing the easy thing. Policy as code. Automated validation. Golden paths good enough that people actually want to use them.

An underrated part of this is testing and security. Automated tests, security scanning, and policy checks in the pipeline aren't extra layers on top of the platform, they're part of the very definition of what it means to deliver correctly. You catch problems early, without anyone having to look over a shoulder.

It gets no less important when AI starts generating infrastructure code for you. Quite the opposite. You can't review everything by hand when AI produces code at scale. Then it's the platform that has to catch the mistakes, not a person. The same goes when AI agents themselves start reaching for tools and systems. Model Context Protocol, the open standard for how agents connect to tools and data, gives a natural place to set exactly those boundaries. The protocol connects the agent to the tools, while what it's actually allowed to do still has to be defined and enforced around it. The idea is the same as with guardrails. The boundaries have to be clear, and they have to be enforced by the system, not left to manual follow-up.

Seen from the platform side, these aren't two different worlds. Whether it's a development team or an AI agent doing the work, the challenge is the same: governance at scale, without sacrificing pace.

It sounds simple. It isn't. But it's what actually scales.