The Three-Week AI PoC That Took Three Months
During this session, I’ll present the different challenges my team and I faced while delivering an agentic AI PoC in a regulated enterprise environment.
The talk will be relevant to techies who can relate to challenges from a development, cloud infrastructure, and enterprise setup perspective.
Main idea: AI cannot always solve the real problem.
Key takeaways: practical do’s and don’ts when delivering an agentic AI solution in a regulated enterprise.
This session tells a short, funny, and at the same time a bit painful story about how a three-week agentic AI PoC turned into a 3+ month project. It is the kind of setup where even the best AI tools cannot save you from the real enemies: enterprise security, private networking, permissions, unclear ownership, and environments that refuse to talk to each other.
In this five-minute ignite talk, I’ll take you from "we have a working demo" to "why is this still not in production?" - including the shortcuts that backfired and lessons that actually mattered.
You’ll leave with practical do’s and don’ts for delivering agentic AI in a regulated enterprise, a clearer view of where AI agents genuinely help, and hopefully fewer surprises when your next "quick PoC" meets production reality.
Lead AI Engineer with 8+ years of experience designing and delivering enterprise AI and GenAI platforms at scale. I specialise in building production-ready AI systems, including RAG pipelines, agentic architectures, and AI platforms that enable rapid use case development across organisations. I combine hands-on engineering with strategic leadership, driving AI adoption, defining architecture standards, and leading cross-functional teams to deliver measurable business impact
LEADERSHIP & COMMUNITY
Mentor AI engineers and data scientists
Speaker at industry conferences (https://sessionize.com/robert-dzisevic/)