Notes on what I'm working through. Mostly enterprise AI, sometimes markets, occasionally something else I can't stop thinking about.
Usage metrics track activity, not whether work actually changed. Design backward from business outcomes, observe real workflows, and treat low adoption as a product signal.
Read on LinkedIn (opens in new tab)Enterprise AI pilots stall when adoption is treated like a software install. Production needs process change, clear owners, and evaluation gates before another agent ships.
Read on Cloudforce (opens in new tab)Shipping agents that hold up in production takes discovery, formulation, design, and evaluation, not only LLM wiring. A structured path from use case to measurable outcomes.
Read on Cloudforce (opens in new tab)Model catalogs grow faster than decision quality. The bottleneck is evaluating outputs against business metrics, not picking a default frontier model.
Read on Cloudforce (opens in new tab)Position sizing and risk limits decide whether a book survives volatility spikes. How to set exposure rules when alpha takes a back seat to staying solvent.
Read Thread on X (opens in new tab)