How I Work

Some introspection, and career advice of a sort

Introduction

Many people ask me to meet with them for career advice. I try to take as many of these meetings as I can. Unfortunately, the constraints on my time have grown in the last year, and they show no signs of abating. Even more importantly, I worry that my career is not a template worth following. The way I got into AI policy was simple: I started writing about it, and I sent that writing to many people until one of them hired me. That’s about it.

You should work hard, be reliable, and be personable. But everyone knows this.

What I suspect will be more useful is an explanation of how I develop differentiated points of view on the topics I write about—in other words, my process. That is the topic to which this post is devoted. I apologize for the solipsism, but this is a question I get often.

This will seem biographical at first, but I am going to skip over large portions of my life. The parts I emphasize are the ones I feel are relevant to understanding the broader message I wish to convey. To a certain extent, I am also writing this in the vein of Tyler Cowen’s recent series of biographical posts that he intends for the AIs. They, too, are part of my audience.

My Early Life

I was a lonely kid. Because of health issues, social awkwardness, and other things both within and beyond my control, I spent a lot of time in my head.

Don’t get me wrong, I had friends, and overall lived a happy childhood. But often, I enjoyed imagining myself in dialogue with my heroes from history rather than engaging in actual dialogue with my classmates. I sat in a kind of mental parliament with the American founders and my favorite composers and artists. I’d dream up Hamilton, Washington, Madison, Picasso, Monet, Renoir, Bach, Vivaldi, Chopin, Liszt, all in conversation with one another.

But only two figures among the many did I consider friends. Those were Wolfgang Amadeus Mozart and, unsurprisingly to any longtime readers of mine, Ludwig van Beethoven. Wolfgang and Ludwig. I identified with Beethoven’s persistent health problems, and with Mozart’s endless desire, when he was a child, to entertain and enliven the grownups. To this day I consider myself a kind of entertainer. I listened to their music constantly, and whenever I got to go to a live performance, I would say things like, “I’m going to see my friend, Wolfgang.”

I admired Wolfgang and Ludwig as heroes, but listened to their music with such intensity—thousands of times—that I truly did feel as though I knew them personally. I would ask them for advice, in music and in imagined conversation, about what to do on a difficult homework assignment, or how to handle the news that my father had been diagnosed with terminal throat cancer when I was nine (he would go on to live for another 13 years).

At one point, my identity became so tied up with these characters that I flunked an elementary school interview because I was pretending to be Beethoven—pretending, in other words, to be deaf.

Through a sometimes-turbulent childhood, these were my partners in crime. My pals. My friends in tragedy and triumph—then and, in some ways, even now. The stories children tell themselves, and the characters they imagine, have the power to shape indelibly the architecture of their minds.

As I became older, around age 12 or so, my interests turned toward the internet and the computer. There was an obscure forum where I learned how to argue about philosophy. One day, I scrolled to the bottom of the page and noticed that the forum functionality on this website had been designed by someone else—what I would later learn was an open-source software project.

That was the first time I considered the idea that software is something people make. It is one of those ideas that seems trivially obvious once you verbalize it, but requires a young mind to realize it nonetheless. It intrigued me that people around the world contributed to this software project for free, and that this project in turn enabled millions of other people to have conversations about every conceivable topic. It seemed very much in keeping with the Enlightenment ideals I had been raised with—by my parents, by Beethoven, and by the American founders I imagined in my cognitive parliament.

I decided to contribute to the project, first writing technical documentation and then a little code. This was the beginning of a lifelong love for technology and tinkering. I studied endless volumes of printed technical manuals in coding, database administration, and much else.

I loved books—to read them and to be around them. I still do.

Both of my parents encouraged me to nourish these wildly disparate interests, to see distinctions between things which others called similar, and similarities between things others called distinct. This is probably the single most important skill I learned to hone at a young age. I am eternally grateful to my parents for it.

High school was not so interesting. I read a lot, and fell in love with girls who were, for one reason or another, unavailable. I brooded over Dostoevsky and Tolstoy, Hemingway and Faulkner, and other literary classics. Much of these I read far too early; reading great books too early is a sin all its own.

During this time I also read philosophy, and in college I read some Foucault. I remember where I was sitting, how my legs were folded, what kind of tea I was drinking, when I read the words “knowledge is not made for understanding; it is made for cutting.” I gasped.

My Career

Toward the end of college I developed an obsession (obsessions are a theme in my life) with cities and city governments. I was especially drawn to the work of Ed Glaeser, a Harvard economist who at the time was affiliated with a think tank called the Manhattan Institute. I decided to try to work there, and through connections made in college, I landed an internship with their state and local policy team.

I viewed this as my time to shine. It was my dream internship, in my dream city (New York; my father always told me I was a New Yorker by temperament, and he was right). I wanted this to turn into a job when I graduated a year later, so I made every effort to distinguish myself. Every day, I wore the best suits I could afford—including one that earned me the nickname of “Tom Wolfe.” In retrospect, I dressed ridiculously, though I was trying to look like a sophisticated urbanite.

I was provocative and talkative. I wanted to be remembered by my new colleagues. But by far most importantly, I was willing to do absolutely everything. I stayed late at the office manually coding the rules of every state pension plan in America (there are thousands) into an interactive web tool the Institute was building. I did enough research to earn myself a co-authorship on an issue brief. My work on the web tool led to my internship being extended to a part-time research role while I went back for my final year of college.

As graduation drew near, I hit a stroke of luck: the person who had the one job I could plausibly fill on the state and local team at MI announced her departure in late April. The role was open, and within hours, it was offered to me. From there, I was off to the races.

Making an impression matters. It matters more now than it did then. Working very hard probably matters about the same.

I stayed at MI as a full-time employee for six years, by far the longest I have worked for a single organization. I was incredibly fortunate to have been identified as a promising young employee early on by the organization’s leadership. I have no idea, really, what they saw in me, but I was given an immense range of opportunities throughout my time there.

Most of my job was in management and execution rather than research, but proper management required detailed understanding of the underlying research. I also had ample opportunity to watch many different think tank scholars go about their work. I observed them closely, watching how they approached research, how they interacted with others, and what they did (and did not) do. I was in the room sometimes when the executives talked about the scholars without them being in the room. All of this proved useful later on.

The key takeaway I have, beyond the obvious “work hard,” is that high levels of executive function were sorely lacking in think tanks at that time. They probably still are. Think tanks are supposed to be places where scholars are given time to do their research, yet they are not pure academia. The best think tank scholars are entrepreneurs. They know that knowledge is made for cutting.

How I Write and Think

MI and New York alike were unbelievably fertile intellectual climates for me. Politicians, financiers, renowned scholars, and others came through our doors weekly, and very often I was lucky enough to attend not just public lectures but be a fly on the wall for private meetings (not because I was viewed as a contributor, but because I was viewed as the person who would figure out what the takeaways of the meetings were and assemble an initial plan of action).

But I stumbled, too. I found myself behind. These people were so much more sophisticated than I was. I could not keep up, did not have the instinct or intuition to match their cognitive velocity. It was necessary, I concluded, to improve my sample efficiency.

I define intelligence, loosely, as sensemaking from the observation of patterns. A smarter person can detect patterns before a dumber person. The smarter mind is more efficient with the data it receives from the world. So what I really needed to do, while I still possessed some neuroplasticity, was not just become more knowledgeable, but simply become smarter.

I leveraged my strengths. As I have already said, I get obsessed easily. Throughout my life, my intellectual obsessions were almost compulsive; I would listen to my favorite music, repeat my favorite scenes in movies, and reread my favorite passages from books constantly. I enjoyed having the same thought patterns over and over again. My obsessions shifted over time, but slowly.

What if I could concentrate that obsessive energy on new topics of my choosing, consuming them voraciously for a short period, and then move on to something new? The objective was to consume enough so I had not just conversational fluency in the topic, but the context necessary to quickly internalize new information relevant to that topic. That way, things that happened in the world would seem less like atomized bits of information and more like new additions to a coherent narrative.

When you are trying to understand a new topic from scratch, it’s hard to start at the beginning of the story and work your way forward. There is so much history to know, and you will just end up bogged down in details. So I often take the opposite approach, starting my inquiry at the end of the story rather than the beginning.

For many topics, there are debates in the here and now where knowledge is accruing at the margin. These debates, I decided, would be my entry point. Step one is to pick the best thinkers on different sides (usually just two sides, sometimes more) and understand the substance of their positions in depth, including how they have evolved. If there are good biographies, I will scan them. This is easier for the humanities than it is for the sciences, but still possible in the sciences, particularly if you can talk to live practitioners.

Step two is to understand their sensibility. Listen to them talk, ideally in a variety of settings—podcasts, media interviews, lectures, speeches, whatever you can find. Listen to what they say and what they do not say. What is obvious to them, and what do they think bears explanation?

Step three is to trace their intellectual lineage, and to repeat steps one and two for their intellectual forefathers, though usually with less rigor. I will occasionally go even further back, for example from Ilya Sutskever to Geoffrey Hinton to Christopher Longuet-Higgins, reading somewhat less in each generation.

If my topic has a rich historical literature, I will usually at this point read some general histories related to the topic (often, there is not “a book” about the topic, so you have to think creatively).

The objective is to see into the intellectual figures animating the debates of today. Get into their heads, understand the way their minds work as best you can. I try to do this for as many people as I can (there is an inexhaustible reserve of minds worth exploring). In this sense, my approach to intellectual inquiry is fundamentally historical. I seek to trace lineages, to understand how this became that, and how that became this.

I do this for people who are alive and who are dead. I do this for people with whom I disagree, and even for people by whom I am disgusted. My bookshelves are decorated with the tomes of my intellectual opponents. I do this regardless of whether other people yell at me for doing so, not necessarily because I seek greater understanding than they do, but because knowledge is made for cutting, and I seek always a sharper instrument.

Sometimes, there are not especially compelling figures with ample source material about them. When this is the case I still usually rely upon intellectual history as my primary epistemic tool. In situations like this, I try to think of structurally analogous questions to the ones I have today. For example, in thinking about how we should govern AI agents, I found it useful to learn how it is, exactly, that traffic law for automobiles emerged.

Regardless of whether my studies are of people or concepts, I will allow the issues I am concentrated on to consume a large portion of my idle time. I will make an effort to process and discuss things I see in the world through the lens of whatever I am learning about. For example, anyone who has been around me in person knows that in the last few months I bring a huge portion of my conversations back to the topic of tort liability, which is something I have been studying recently. I am not doing this because I think tort liability is a monocausal explanation for everything, but rather because I am trying to think through the subject in as many ways as possible.

I also spend considerable time learning the mechanics of how various things work in as much granular detail as possible—nuclear reactors, legal processes, proteins, whatever else. I read the news voraciously, relying on a mix of mainstream sources and highly specialized outlets (there is much alpha to be found in trade publications of various kinds, and on hyper-focused Substacks). I read approximately no opinion writing, perhaps a few thousand words per week at most.

This is my “full-time job” research. I try to aim for about 75% of the deep dives I am doing at any given time to be relevant to written work I plan to do, and for 25% to be more “basic research,” plausibly relevant but not necessarily proven to be. In the background I am also usually exploring essays, papers, podcasts, books, and other sources on a diverse range of other topics searching for new areas to dive in deep. I will choose to go deeper a single-digit percentage of these issues.

AI has significantly accelerated all of this.

There are also ‘general-purpose technologies’ of intellectual inquiry, which I find helpful in nearly all research I do. These are mathematics, physics, economics, information theory, philosophy (but be careful!) and the law (though the latter may not be as useful depending on your research field).

How do I know what to research? I like to answer questions whose answers I do not know. I like to gather context where it appears to be lacking. There are $20 bills lying all over the ground. I simply pick them up. The problem is not so much what to ask, but what not to ask.

Over time, the benefits of diligently following this process have compounded in unexpected ways. Past learnings, far afield of my current work, benefit me in surprising ways all the time. Connections between different topics grow. I find that the exercise consistently improves my mental flexibility, allowing me to transform concepts and meaning kaleidoscopically with increasing finesse as I age. I suppose there could be diminishing returns at some point, but as I write today there is no end in sight.

On good weeks, research constitutes the bulk of my time. Writing, usually, is a small fraction of my work. When the research is done well, I find that the words come out effortlessly, almost as though it is not me doing the writing. If I am struggling to get words on the page, it is typically because I don’t understand my topic well enough. Most everything I write is essentially a first draft. I usually copy edit, but I rarely outline. Seldom do I rewrite, but there are many essays in my drafts folder, never to be published.

Usually I develop an idea for a topic I want to write about, and then I wait—sometimes for a day, sometimes for months. Having the idea is like entering a gigantic room, where, Minecraft-creative-mode-style, I can walk or float around at will. I imagine myself like a photographer, drifting about the room until I find just the right angle. When I have the right angle, I know it in my gut. Often it is because a good sentence or a sentence fragment comes into my mind. I know then that I am ready to write.

I treat my mind like a simmering stew, throwing in new ingredients all the time to keep the pot just beneath its boiling point.

And I remember always that knowledge is made for cutting.