Conway’s Law in AI
Melvin Conway’s observation that a system’s design mirrors the communication structure of the organization that built it, remains valid in a world dominated by AI. LLMs and autonomous agents do not escape organizational gravity, they actively amplify it.
This continuity exists because AI outputs are built on the foundational data, human biases and organizational silos embedded during the training and prompt-engineering stages. When an enterprise trains models on internal repositories, the data carries the historical boundaries, terminology gaps and information silos of the source departments or teams. Similarly, when different teams develop prompts in isolation, they implicitly include existing logical silos into production environments. This structural mirroring becomes most evident in multi-agent architectures. Organizations design AI workflows to mimic traditional hierarchies, flows and processes. Consequently, the automated workflow inherits the exact bottlenecks and communication characteristic of the human hierarchy it copies.
The operational reality remains: an enterprise cannot sucessfuly adopt AI faster than it aligns its own decision flows, language and cross-functional teams. AI does not correct broken communication, instead it automates and accelerates it. If human teams lack a unified language, the AI tools instructed to generate specifications, write code or manage operations will inevitably execute that same misalignment. Achieving cross-functional AI requires standardized data structures, shared corporate taxonomy and operational alignment before technical development begins. System design remains linked to organizational reality, making Conway’s law more relevant than ever.
Transparency as Value Generator
Indirect Prompt Injection
Yes Man AI Agent
Sandbox Paradox