Agent stacks keep rebuilding the sensor-fusion pipeline from scratch, without the pieces that made it work. Priors, calibrated gates, coordinate transforms, and recovery are not new ideas — but every one of them is missing from most agent architectures today.

Deloitte's Tech Trends data shows 93% of enterprise AI spend goes to technology and tooling while just 7% funds culture, change management, and learning. Bill Briggs argues this imbalance directly explains why fewer than 30% of agentic pilots reach production at scale.

Stefano Fiorucci trained a small open-source model to outperform GPT-5 Mini at tic-tac-toe using reinforcement learning with verifiable rewards. The key lesson: environment design—reward signals, opponent calibration, batch sizing—determines whether RL training succeeds or collapses.

Enterprise agents struggle less from weak models than from human-shaped interfaces, raw observability data, and unsafe workflows. Andre Elizondo argues the real work happens earlier: transform the data, constrain the tools, and build evaluation loops that make production decisions inspectable and safer.

Andrej Karpathy rode in a near-perfect Waymo demo in 2014. It took a decade to become a paid product. That demo-to-product gap—not model architecture—is the binding constraint across self-driving, humanoid robots, and AI-powered education.