Lawrence Shieh

As AI agents become more capable, the biggest unanswered question remains: how close are we to 1–2% error rates that would make them truly reliable? Despite breakthroughs, we’re not there yet, and the path forward involves more than just better models. This panel breaks down the technical, cognitive, and operational barriers to high-accuracy agentic AI, and what needs to change in how we build, evaluate, and deploy these systems responsibly.

• Achieve trustworthy AI performance by defining and measuring accuracy beyond benchmarks
• Close the reliability gap by understanding what current models can and can’t do autonomously
• Set realistic expectations by distinguishing between narrow AI success and generalizable agentic behavior
• Improve outcomes by designing AI agents with safety, feedback loops, and human oversight