This work concerns the space between information and the people who use it. Information often arrives in the form its system produced rather than the form its reader needs. Given sound structure, the same content can take any number of forms, so the central question becomes which presentation best serves a particular reader, task, and context.

Principles

• One structure, infinite presentations. Presentation is where people meet information, and it should never be locked into a single form. When content carries explicit structure (headings, relationships, data, and meaning), it can become a page, a slide, a canvas, a chart, audio, or a summary, and each form can be shaped to the person using it. Structure is what makes that range possible, and presentation is where its value is realized.
• Design for the work people already do. Change lasts when it fits existing routines: the documents people already write, the tools they already open, and the decisions they already make. The aim is to make the better practice the easiest one, rather than to add a new obligation on top of the old.
• Measure against real use. An evaluation matters only when it reflects the conditions in which a tool will be used: real documents, real tasks, real people, and real limits of time and hardware. Benchmarks are a starting point. The test is whether the work gets done, and the limits of every result are reported with it.

Much of this work is carried out with AI models running locally. Those models are evaluated on the tasks they will actually perform before they are trusted.

Explore

• SkillsWorkflow skills for agent-assisted work
• ProjectsInterfaces, tools, and pilots
• TalksA trilogy on the human interface
• ConceptsInfinite canvases and drill-down images
• EvaluationTesting local AI models

Essays on agents, evaluation, and accessibility are on Writing.