Building an AI risk agent at Gusto required close collaboration between engineers and domain experts, not just technical implementation. The team discovered that success came from tight feedback loops with operations specialists who understood the nuances of risk assessment, pragmatic architectural decisions favoring learning over purity, and treating prompts like code specifications. The resulting system reduced manual reviews by 70% and achieved 6x faster approval times by combining AI automation with human oversight safeguards.
Table of contents
The Myth vs Reality of AI at WorkFrom POC to ProductIt Takes a VillageThe MessGet Xao Yang ’s stories in your inboxImpact and What We LearnedSort: