Bylaw

About

Bylaw is a check that runs before an AI agent writes to your systems. If the agent is about to act on evidence that is wrong, outdated, or don’t agree across sources, we block the action or send it to a human first. We all met our first year at the University of Waterloo studying computer science. We have previously interned at Meta, Optiver, and BitGo, places where making decisions on unverified data could have lethal reprecussions. We kept running into the same failure deploying agents: the agent was allowed to call the tool, but nobody had verified the facts behind the call. Guardrails ask “can this agent use this tool?” Evals catch mistakes after the fact. Neither checks whether chat, docs, CRM, and tool results actually agree at the moment of the write. That’s why the same refund or CRM mistake keeps coming back even after you fix the prompt. This is hard because evidence lives in five places at once, goes stale between lookup and action, and every write needs different proof. It hasn’t been productized because agent teams ship fast and patch with prompts; platforms build one-off checks instead of a reusable gate in front of every sensitive tool call. The same way finance built approval gates before money moves, every company giving agents write access will need this layer.