Es Devlin is building a living portrait of the entire United Kingdom, pulling faces from everyday people into London's National Portrait Gallery. This week, researchers published PREPING, a framework for building AI agent memory without predefined tasks, letting agents accumulate context the way a person accumulates experience. Both projects are doing something structurally identical: trying to construct a coherent identity from distributed, unscripted fragments. Neither can fully control what they get.
Distributed Portraiture and the Coherence Problem
Devlin's project is radical because the National Portrait Gallery has historically been a place where subjects are pre-selected, curated, and legitimized by institutional judgment. Opening that frame to everyone doesn't dissolve the institution; it redistributes the editorial problem. The same tension runs through the PREPING paper, which notes that most agent memory systems rely on curated demonstrations or structured post-task reflection. Building memory from ambient experience, without task scaffolding, produces richer but less controllable recall. A 2024 paper in Nature Human Behaviour by Molly Crockett and colleagues on identity coherence found that people construct continuous selfhood from fragmentary memory through active narrative editing, not passive accumulation. Devlin is doing that editing. PREPING is trying to automate it. Neither is entirely succeeding, and both are interesting precisely because of that failure.
The Archive as Agent, the Agent as Archive
The deeper connection is about what it means to represent a collective. Devlin's portrait of the UK and the two-dimensional AI agent design framework published this week both grapple with the question of topology: how do you structure something that is simultaneously individual and distributed? The agent paper distinguishes between cognitive function and execution topology, which is a formal way of asking whether an agent thinks alone or in concert. Devlin's installation asks the same thing about nationhood. In both cases, the answer involves accepting that the portrait will always be incomplete, always updating, and always contested. That is not a flaw in the design. That is the design.