About

Anton Braverman

I work on the unglamorous half of enterprise AI: not the demo, but what happens after it ships. Taking agents to production and holding them to what the business actually intended. In practice that means the gateway in front of the models, the guardrails and content safety around them, the observability that makes their behavior legible, and the evaluation that decides whether they can be trusted at all.

Most of what breaks at that stage is not a model problem. It is an organizational one: an install that no one decided the purpose of, a control that was assumed rather than enforced, a judgment call that never got written down. A lot of my writing circles that gap between what a system was meant to do and what it was actually permitted to do.

Most days the work is hands-on: enterprise AI agents going to production across AWS, Azure, and GCP. This site is the quieter counterpart, a place to think out loud without a pitch attached.

I try to keep these notes honest about their own edges. Where something is well-supported, I say so. Where the right architecture is still an open question, I say that too. If you have read something here and want to argue with it, that is usually the best kind of message to get.