Designing for the other reader

In this guest workshop for Xinran’s Maven cohort, I explored a new audience for design: the AI agents interpreting the systems beneath our interfaces. These agents read prompts, knowledge bases, brand kits and documentation before shaping what users eventually see.

I focused on three neglected design surfaces: context, recall and trust. We examined context dilution, progressive deletion and how products can make memory legible by showing what an agent knows, what it has dropped and why.

The workshop also reframed trust as an output-level question: who owns this information, when was it last verified, and is it sufficient? I showed how designers can trace agent failures back to stale sources, unclear instructions and missing context.

The central takeaway was simple: information architecture now serves machines as well as people. Designers must shape the substrate beneath the interface, and ask in every critique: will an agent read this, and how does the design serve it?