AI cannot optimize a company it cannot understand

Details

This Fast Company opinion piece makes an architectural argument about why AI cannot yet optimize most organizations. The author draws a sharp distinction between context—information fed to a model at runtime—and a true organizational data model: a formal representation of identities, relationships, permissions, constraints, and valid states. Companies confuse the two at their peril. The core claim: even capable LLMs are useless optimizers if they cannot formally represent the organization—its customers, dependencies, bureaucratic processes, risk tolerance, and cascading effects of change. Cramming more context into a prompt is not a substitute for a structured ontology. The proposed solution resembles what Palantir has sold for years: an operational layer representing real-world entities as objects (factories, orders, workers, customers) and actions as verbs (adjust price, change distributor, execute workflow). This digital twin of the organization is the prerequisite for AI to reason about systemic optimization rather than isolated tasks. The catch the author acknowledges: building ontologies is expensive and labor-intensive, which is why Palantir deploys forward-deployed engineers (FDEs) on-site and why the economics of that model are substantial. The piece is diagnostic rather than prescriptive—it correctly identifies a real gap in enterprise AI but stops short of naming vendors automating ontology construction or alternative cheaper approaches.