Descriptions:
Jack Dorsey published a blueprint last week for AI-powered “world models” — software that maintains a continuously updated picture of everything happening across a company, from what’s being built to where customers are struggling, so that nobody needs a middle manager to shuttle context between the people doing work and the people deciding what work to do. The post got 5 million views in two days and immediately triggered enterprise software vendors to rebrand their products around the concept. This video from Nate B. Jones takes the idea seriously but goes well past the pitch.
Jones breaks down three distinct architectures that companies are currently building under the “world model” label: a conversational LLM layer that ingests all company data and answers queries in natural language; an ontology-first structured graph that only represents pre-defined, categorized relationships; and Dorsey’s own “signal fidelity” approach at Block, which is built around high-quality transaction data on the theory that money is inherently honest. Each architecture has a different, specific failure mode — the LLM layer hallucinates causal relationships, the ontology is precise but blind to emergent patterns it was never designed to capture, and the transaction-signal approach creates an illusion of interpretive authority because the input data is so clean. The shared blind spot across all three: they handle information flow well and judgment not at all, and the line between the two is never clearly drawn.
The video grounds these failure modes in real organizational cautionary tales — Zappos’s Holacracy collapse, Valve’s undocumented power structures, Medium’s operational struggles — and closes with a practical readiness framework for organizations evaluating whether and how to build world models of their own.
📺 Source: AI News & Strategy Daily | Nate B Jones · Published April 19, 2026
🏷️ Format: Deep Dive
