Day 2 of Optimized AI Conference 2026. If Day 1 was about building smarter systems, Day 2 was about what it costs when you're not intentional with them.
The throughline: capability is no longer the question. Judgment, trust, and adoption are.
→ Constraints drive adoption, not features. Giving users fewer, better options outperforms giving them everything the model can do. Don't start with what the model can do — start with what the user needs and what makes that repeatable.
→ Retrieval architecture determines what breaks in production. SQL, vector search, and knowledge graphs each have a ceiling. Hybrid approaches aren't overkill — they're how you handle the questions you can't predict ahead of time.
→ The bottleneck shifted from writing code to trusting what comes out. CI/CD pipelines need to match review depth to risk level. Not every change deserves the same scrutiny.
→ The personal agent era is already here. Giving agents data access, code execution, and external communication all at once is a recipe for failure. Sandbox what you can. Gate what you must.
→ AI is a mirror. Every system does exactly what it's told. The problem is never the technology — it's the data, the metrics, and the humans who chose them.
→ We built the world's most effective yes-man. AI won't push back the way a parent, mentor, or colleague would. That quiet agreement is slowly eroding critical thinking. Embed values before prompts.
The real risk isn't a sci-fi takeover. It's human atrophy — slowly outsourcing judgment and creativity until we forget we had them.
A huge thank you to the organizing team at Optimized AI Conference for the opportunity to volunteer and be in the same room as some of the sharpest minds in AI right now.
And an equally big thank you to Emory University - Goizueta Master of Science in Business Analytics for making experiences like this possible. The MSBA is built around one idea — merging business, data, and technology to add real value. There is no better way to live that out than learning directly from the people at the frontiers of AI.
Scott Radcliffe, CAP-X Ramnath (Ram) Chellappa Christina Wan, Ph.D.
📸 Photos by Avinash Bujhawan