Fifteen years on one question

I sat down recently to look at what I've actually been working on for fifteen years, and the thread surprised me. The projects looked unrelated from the outside, and they had felt that way from the inside too. But each of them, when I lined them up, turned out to be an attempt to answer the same question.

The question, more or less: how do you make a digital thing — or an idea, or a body of knowledge, or a conversation — tangible enough that someone can hold it, work with it, and make a decision from it?

Five rooms, same question.

Tangible user interfaces. I started on this in graduate school in Germany, writing a thesis on what makes a tangible interface work as an experience. The premise was that the screen isn't the right interface for every kind of thinking; for some, you want to grab a thing with your hands, move it, watch what happens. The question was how to give a digital concept enough physical form that someone could think with it directly. Some of the answers are still with me: a wooden dome that controlled a digital map of the MIT campus, a piece of clay you sculpted to see how a landscape would behave when it rained, a glass eyedropper you could fill with a file from one screen and pour onto another. Most of those interfaces didn't survive the move from research lab to anywhere useful, but reading the field taught me where to look. Almost all of it came out of one place: the MIT Media Lab. That's the start of an admiration I've kept since.

Semantic web technologies. The PhD continued the move into knowledge: semantic web technologies, and smart environments where structured data shaped how a space responded to the people in it. Different scope, same move underneath. Instead of making one digital object physical, the question became how to give knowledge itself a shape, so that the shape would derive new information. Knowledge structured as a graph isn't only navigable; it's generative. The relationships in the graph let you infer things you never wrote down. Make the structure visible, and you can reason over it. Make it invisible, and the same knowledge is inert.

Data visualization. Then years of work on visualization as a practice, starting during the PhD with D3.js when the library was new and Mike Bostock was the genius everyone in the field watched. I've followed his work since; his company Observable is still where I reach when I need to wrangle and reason over data. At Thomson Reuters the practice landed on two problems. One was products that ran analysis over huge volumes of news data to advise traders, where the design question was how a trader could trust an AI-powered recommendation. The answer turned out to live in the UX more than in the model: sourcing the "why," making each piece of evidence traceable, putting the structure of the reasoning where the trader could see it. The other was a vast but rigorously structured knowledge graph for financial risk, visualized at scale, which brought me back to thesis territory through a different door. Two domains, one move underneath: give a body of evidence a form someone can think with.

Smplrspace. Then the spatial chapter. Buildings carry an enormous amount of data: equipment, occupancy, energy use, leases, work orders, layouts. Most of it sits in rows of records that mean almost nothing to the people who actually run the space. The question was how to make building data tangible by anchoring it to the space it describes. Show the equipment on the floor plan where it sits. Show each lease as the rooms it covers. Show the work orders in the room they happened in. The data doesn't change. Its legibility does. Decisions about a space become possible because the information is finally in the shape of the space itself. Smplrspace runs on two surfaces: a platform real estate teams use directly across portfolios, and a developer platform that other software companies embed when their own products need spatial intelligence on top.

The Mesh. And now the chapter I'm in the middle of, built on the side of Smplrspace and powering our Claude experience across the team. The question is whether thinking itself can be made tangible, specifically the thinking that happens between a person and an AI. An AI conversation is information-dense and structurally invisible. You finish a session and the understanding evaporates with the chat window; the next session starts blank, the team has no shared view of what was worked out. The Mesh is what happens when you decide that thinking deserves an artifact: a permanent, addressable, visible page the team can return to and the next agent can load. A substrate, so AI-collaborative thinking can compound across sessions instead of resetting with each one.

The Mesh paragraph above is a sketch. Each of the ideas in it deserves a post of its own: what a thinking artifact actually is, what cognitive debt looks like in practice, what an agent-native team needs from its substrate. They're coming.

Five rooms. One question, scoped to the medium I happened to be in.

This is what I mean when I talk about visualization, and why I think the word is usually used too narrowly. It isn't a presentation layer at the end of a process. It's a thinking layer at the start. The form is part of the meaning. The choice of what shape to give a thing is upstream of every decision someone will make from it. Most domains treat visualization as the polish step. It is actually the gate: whether or not a decision can be made well depends on whether the thing has been given a form your mind can read.

The same is true of design. Design isn't decoration applied at the end; it's the judgment about what should exist and what shouldn't. Visualization and design aren't adjacent crafts. They're the same instinct, working in different materials.

I didn't realize for most of those fifteen years that I'd been answering the same question in different rooms. The recognition was recent, and it surprised me. It also explained why the projects I cared about felt connected even when nobody else could see the link, and why I keep ending up building visualization systems regardless of what the title on the door says.

Most careers look more coherent in retrospect than they did at the time. The thread isn't visible while you're on it. It's worth stopping to look.