Intro

In this article:

• my one-hour VoxDev talk on how economists should get started with agentic coding
• some of the reaction to Alexander Kustov's great work thinking through the agentic coding revolution
• some of the reaction to Andrej Karpathy's vibe-coded AI jobs exposure project

Let's get into it.

Briefs

A Free Intro to Agentic Coding for Economists

I gave a one-hour VoxDev talk laying out my view, for an audience of development economists on a low budget, on how they should get started with agentic coding. Big thanks to Oliver Hanney, who organized the whole thing and made sure it ran really smoothly.

The Zoom room overflowed, so a lot of the live viewers had to go to YouTube instead. There were about 750 live viewers in total.

If you want to watch it, the talk is on YouTube. You can also find the slides and the written version on my site.

Alexander Kustov and the Academic AI Backlash

Alexander Kustov had really good reflections, both in his Twitter thread and on Substack, on the agentic coding revolution and its implications for social science. He's a political scientist at Notre Dame, but I thought what he had to say was plenty applicable to economists as well. (part I) (part II)

What was striking, though, was the reaction. Kustov engages a lot with social scientists, especially political scientists, on Bluesky, and their views on the utility, importance, and ethics of AI are, in my view, basically insane. I find the mentality genuinely hard to understand.

One example is Emily Bender, an academic who is known for particularly bad takes on AI, and who went so far as to attack Kustov's tenure case directly by suggesting he was not actually tenure-worthy. Nothing she says is credible, and I don't think it has any real effect on him. But it does point to a more important question: how is this kind of professional hostility to AI going to affect its take-up, and especially the impact it has on students?

Karpathy's AI Exposure Project

Andrej Karpathy was on the founding team at OpenAI and is the guy who coined the term "vibe coding." He posted a vibe-coded project on Saturday night where he took occupational data from the Bureau of Labor Statistics website and created a treemap visualization of AI exposure by profession. He sent the BLS descriptions of each profession to an LLM API and asked it to assign a number from 0 to 10 for that profession's AI exposure. (repo) (live demo)

The internet found this and then went a little crazy with its interpretations of what was, in reality, a simple visualization Karpathy put together quickly to work through some ideas. But it ended up instigating a useful discussion, with economists providing a corrective to some of the worse interpretations. (Kaito tweet) (Kevin Bryan tweet) (Alex Imas tweet)

I also recorded a short video on the episode that walks through the repo and the prompt in more detail, and makes the same basic point: AI exposure should not be read as displacement.

I also wonder what the economics profession's own exposure to AI is. It seems quite significant. But the harder question is what that is going to imply for the need for economists.

The short- and medium-term impact on the software engineering market is already non-trivial, especially once the profession starts to adapt more significantly to AI usage. I know programmers who still do not use agentic coding tools. But at least in the short and medium term, as agentic coding tools unlock new areas where it becomes economical to have software, there may also be an increase in demand for programming.

It is possible that something similar happens in economics. My main concern for the future viability of economics job markets is really the employability of economics undergraduates, both in terms of the signaling value of the degree and in terms of what is actually taught in an economics undergrad.

In the United States, it is pretty well known that you should not do an economics undergrad if your goal is to get a PhD in economics. That is a bit of an exaggeration, but there is some truth to it. Economics is maybe uniquely a field where you can say something like that and not be totally wrong. The reason is that the economics PhD is much more technical than the undergrad, while the undergrad has often been treated at many universities as the easiest well-paying major that exists.

You graduate, then get some policy job running regressions and making reports, or you become a consultant and start out at something like $70,000 to $80,000 a year right out of undergrad.

But I no longer know that this is still the case, or that these jobs are still there in the same way. If that stops being true, then people will stop doing economics undergrads. And to the extent that the econ job market is, at some fundamental level, tied to that demand, what does that mean for future PhD job markets?