Real-world AI failures — and what they teach us about building systems that actually work.

Everyone loves to showcase AI success stories — but the truth is, most AI projects fail. Quietly. Expensively. Repeatedly.
Models that never reach production. Chatbots that frustrate users instead of helping them. Automation pipelines that break under real-world complexity.
Over the past three years, as an AI builder working on more than 20 AI projects — from healthcare and chatbots to banking and industrial automation — I’ve seen firsthand how even well-funded, well-planned systems can fail in unexpected ways.
In this talk, I’ll share real AI use cases that didn’t go as planned — what failed, why it failed, and what we learned in the process.
From misaligned objectives and brittle prompt engineering to missing data context and weak orchestration between human and machine workflows — every case reveals a deeper pattern behind why AI underperforms when it leaves the lab.
This isn’t about pessimism — it’s about engineering realism. You’ll see how failure analysis can become a design tool, helping teams build AI systems that are reliable, auditable, and adaptable to change.