I work at Sustainalytics. From the outside, this work often gets reduced to "environmental ratings" - companies taken in, companies scored, a green or red number. In reality, its core is a less spectacular word: materiality. To assess materiality is to decide, for a given company, which factors actually move its risk and its value - and which are just noise. Not everything good or bad counts the same. The work is to separate what matters from what only seems like it should matter.
Materiality is, at bottom, the discipline of deciding where it's worth looking.
And sitting inside this machinery day after day, I started to see something you can't see from the outside. What truly produced the transition toward a lower-emissions world was not, first and foremost, virtue. It was legibility.
Thirty years ago, carbon was an invisible risk: diffuse, spread across decades, impossible to put in a spreadsheet. No one could allocate it, because no one could measure it. What has happened since was not just green technology - it was the slow construction of an apparatus for reading. A common unit, CO2e, that can hold anything. Taxonomies. 1.5- and 2-degree scenarios. Reporting standards. Risk ratings. Thirty years of coordinated work to turn an abstract threat into something measurable, comparable, allocatable.
Only once carbon became legible did the trillions start to move.
And here I had the click that started this essay: I do, for a living, exactly the thing the transition to AI needs most and doesn't yet have.
Because the transition to a world with far more AI is, right now, exactly where the climate was back in the '90s. Pre-legible. We all sense something big is coming, but we don't yet have the apparatus to read it. Evals, model cards, "AI literacy," safety frameworks - these are the first sketches of the instruments, not the finished instruments.
And the difference from carbon is even more uncomfortable. With carbon, at least we agreed on where we were going. "Net zero" was the destination; the argument was only about the route. With AI we don't even agree on the destination. "More AI" is not a goal the way "less carbon" was a goal.
That's exactly why I want to be clear, because it's easy to slip here: I'm not saying "more AI = good, the way less carbon = good." That symmetry is false and falls apart in a single move. What transfers from one transition to the other is not the virtue, but the method. A world-scale change doesn't become steerable because it's good or bad. It becomes steerable when it becomes legible. Until then, it simply happens to us.
So when I say "it's worth investing attention in AI," I'm not reaching for a warm metaphor. I'm using the technical term of my trade. Investing attention is materiality: deciding what truly matters and what is noise.
And the question that, for carbon, took thirty years and thousands of people to answer - what is material here? - is, for AI, still wide open. Which capabilities actually move risk and value? What is hype and what is structural? Which company has integrated AI deeply, and which just glued a chatbot onto an old process? These aren't philosophical questions. They're exactly materiality questions, asked about the second transition.
And it connects, for me, to something else I've written. There I was talking about autonomous agents that, left to run, develop their own trajectory and break away from the intent you started them with - and about how, on a personal scale, I prefer to stay present, with a partner agent, so that what comes out at the end is recognizable as mine.
Legibility is exactly the same instinct, but at civilizational scale. An illegible transition is like an autonomous agent that has drifted: something that happens to you, not something you steer. Making it legible means staying oriented - still being able to point and say that's where it's worth looking.
The same instinct, at two altitudes: staying present, and staying oriented.
And if I look ahead, this isn't just an observation, it's also a road. The work of making a transition legible doesn't end in a report. A report is a photograph. Legibility is something else - the ability to keep seeing while the ground shifts under you. That's why I say road, not destination: you don't make a transition legible once, you keep it legible. With carbon it took us a generation just to open our eyes. With AI, we're only beginning to learn how to look.
And it all starts from the same place: you choose where you look.