Conway's Law Reexamined: Why AI Agents Don't Fit the Org-Chart Mirror Model
Conway's Law holds that a system's structure mirrors the communication graph of the teams that built it, with interfaces forming wherever human coordination must cross boundaries. A developer essay argues this model does not cleanly extend to AI agents, because agents lack persistent, project-specific memory and reset their context each session rather than accumulating shared priors over time. The author frames communication in terms of transfer speed and shared priors, concluding that agents face an equivalent of 'context pollution' in multi-agent setups, much as human teams suffer from contradictory or siloed communication. Human-agent collaboration is further bottlenecked by the pace at which a human can read and act on agent output, making the reading and comprehension step the true constraint to optimise. The essay warns that marketing AI as a seamless 'teammate' remains premature, since the interface between humans and agents is still far less settled than decades of human-to-human collaboration norms.
