Donald Hruska is an engineering manager at Vanta, a co-founder of Draftbit, and a former engineering lead for Retool Agents. He builds tools that make sophisticated software easier to deploy inside organizations with demanding security and operational requirements.
Hruska studied computer engineering at the University of Illinois Urbana-Champaign and became a technical lead at automotive-technology company DRIVIN before co-founding Draftbit, where he served as vice president of engineering. The company joined Y Combinator’s Winter 2018 batch and developed a visual mobile-application builder that gives developers access to their generated source code. Its open-source react-native-jigsaw library supplied reusable React Native components and theming.
At Retool, Hruska led new-product engineering across mobile applications and applied AI before overseeing its agent product. His approach to enterprise automation centers on several practical distinctions:
Bounded agent execution: Agents operate through repeated reasoning, tool calls, and self-verification; explicit iteration limits prevent runaway loops and uncontrolled model costs.
Production-ready enterprise agents: Integrations must respect authentication, permissions, auditability, compliance, and access to sensitive systems. A working prototype does not resolve those operational requirements.
Agent evaluation and observability: Teams need to inspect individual executions, model behavior, token consumption, runtime, and estimated costs before entrusting agents with consequential workflows.
Build-versus-buy for agent infrastructure: Companies should consider building agents that define their competitive advantage while using managed platforms for supporting processes when existing connectors, security controls, and monitoring reduce engineering overhead.
In his AI Engineer World’s Fair talk, Hruska connected the effectiveness of coding assistants to a broader enterprise opportunity: giving models controlled access to real business systems while preserving human accountability.
A small tool-calling loop can power an agent, but enterprise value depends on permissions, evaluation, observability, and choosing which systems deserve custom engineering.
Where does the infrastructure investment become useful work?
▶ Watch ↗