Sumaiya Shrabony is a technical program manager and open-source developer who builds safeguards for AI systems that can produce convincing work without meeting basic standards of accuracy, originality, or accountability. Her projects apply software-delivery discipline to agent workflows and expose unreliable enterprise data before organizations deploy AI assistants.
Originally from Bangladesh, Shrabony studied computer science at Hofstra University and information systems at the University of Colorado Denver. Her career has included analytics and operational work at Optimizely and responsibilities involving data infrastructure, access governance, and AI adoption at the University of Colorado Denver, where she works in the Department of Computer Science and Engineering.
Her open-source agentic-content-system coordinates research, planning, drafting, verification, and review through 19 specialized Claude Code skills. Shrabony initially built the system with 16 skills; as it expanded, seven workflow handoffs created opportunities for polished but defective material to move downstream unnoticed. Her AI Engineer World’s Fair presentation distills those failures into practical engineering requirements.
- Blocking production gates. Agent workflows need regression tests, scheduling alerts, schema validation, staging boundaries, and audit trails. Shrabony recommends protecting the most consequential handoff first with a gate that can prevent invalid work from advancing.
- Polished failure detection. Plausible output can obscure generic writing that loses its author’s voice, unsupported factual claims, and recycled ideas. Her safeguards combine voice contracts, claim verification, deduplication, and records identifying precisely where validation failed.
- Semantic readiness for enterprise AI. Her DataReady project uses Gemini to diagnose problems in the semantic layer behind business-intelligence systems, including ambiguous metrics and unreliable analytical foundations that can undermine an AI copilot.
- Human accountability and consent. In her Ground Truth newsletter and writing on workplace automation, Shrabony distinguishes tasks AI can draft from decisions people must verify and own. She also advocates clearer training-data consent and default protections against unwanted participation in AI systems.
