about
The engineer behind the system.
One move runs through every part of this site: build the rule into the structure of the system instead of trusting it to be remembered, so that breaking it becomes impossible rather than merely discouraged, and keep one human at the gate on every action that cannot be undone.
the approach
Three principles, one move.
The same three commitments recur in every pattern on this site. Together, they are the method: prefer structure to policy, pay the cost up front, and keep a human on the irreversible step.
- the move
Policy becomes structure
A written rule depends on every future reader remembering it; a capability that was never granted depends on nothing. Wherever a guarantee matters, I convert it from a rule that can be talked past into a capability the system simply does not have. The safety property stops being something to remember and becomes something the system cannot violate.
- the cost
Friction paid up front
Every guardrail is a small tax at authoring time: two files kept in agreement, a question answered before an agent can grow, a deviation written down rather than slipped through. I pay that friction deliberately, because it buys a whole class of silent failure made impossible downstream, where it would otherwise surface as an error no one catches in time.
- the gate
A human on the irreversible step
Where impossibility is the wrong tool and judgment genuinely has to happen, the answer is never "automate it anyway." Routine, reversible work is driven down to one command. The one action that cannot be undone stays behind an explicit human decision.
About the maker, in person
the maker, in person
R Montufar
Systems & AI-workflow engineer, with a cloud compliance-automation background
My work started with compliance automation: taking policies that lived in documents, cloud data that lived across systems, and audit checks that depended on memory, then turning them into repeatable workflows.
That shaped how I think about AI agents. I do not see them as magic. I see them as systems that need structure, context, limits, and a clear path for escalation. The safe version of an agent is not the one that acts the most independently. It is the one that knows when to stop, explain, and bring a human back in.
This site documents that journey. The framework I’m building is about making AI workflows more reliable by design: scoped inputs, structured reasoning, tool-aware execution, clear checkpoints, and no irreversible action without human approval.