Thread by @johnarnold

After many conversations over past year with friends, business associates & policymakers about the future of AI job disruption, I’ve tried to get my thoughts in order. With the caveat that I have no specific AI expertise, here they are. Comments and corrections encouraged.🧵
1/n

AI coding tools crossed an important threshold over the past year with release of Anthropic's Claude Code. Previously, largely sophisticated autocomplete systems that created code snippets, they’re now agents capable of completing substantial engineering tasks.
2/n

Developers were shocked at how good the tools were. Those previously skeptical of AI coding did a 180. Very quickly, the role of many software engineers began to change, shifting from writing code themselves to supervising, directing, and reviewing the work of coding agents.
3/n

AI coding tools had been improving for years, and adoption was growing, but it was slow. The tools were useful but limited. For specific tasks they could save time but that didn't generalize across broader workflows. That changed when the tech crossed a critical threshold.
4/n

With a single release cycle, AI coding became good enough to be broadly useful, and adoption skyrocketed. Anthropic's preferred revenue metric exploded from an annualized $3 bln in May '25 to $47 bln one year later, historic growth never seen in any business before.
5/n

Many tech firms responded with layoffs in 2026, some specifically citing AI coding. There is debate about how many jobs AI has actually displaced. Some companies were already overstaffed and their CEOs gained social cred within tech circles by claiming to be early adopter.
6/n

Some companies were trying to become leaner to accelerate employee adoption of AI. Others were shifting capital toward AI. And there has probably been some genuine AI-driven job displacement. Regardless, nearly every tech firm believes much larger disruption is coming soon.
7/n

Those in coding are freaking out. If models improved this much just in past year, what does next year look like? There is high confidence that AI will structurally disrupt the job of coding. There is active debate about the # of future coding jobs. Jevon's Paradox yada yada.
8/n

The bigger question is whether software is merely the first domino. If models crossing the threshold of "good enough" created this much disruption in coding, which fields are next? AI maximalists, who are increasingly influential in Silicon Valley, believe the answer is many.
9/n

In their view, substantial white-collar job displacement could arrive in the near term, with blue-collar disruption following as robotics improves. This is where the debate gets interesting.
10/n

There are several reasons software was likely disrupted first:
1) The work is already digital and performed through text
2) Coding has tight feedback loops that allow for easy testing of whether the AI output worked
11/n

3) Massive training datasets exist through gajillions of lines of publicly available code
4) Tech workers are natural early adopters of new tech tools
12/n

5) There is little real-world friction to adoption in software, like regulations
6) Software employees are highly paid so there is strong incentive to automate their work
13/n

Taken together, software may be close to the ideal environment for AI adoption. The key question is whether other professions share enough of those characteristics for a similar pattern to emerge.
14/n

Though professions like customer support, translation, and some back-office share some of these characteristics, most do not. Will AI-driven disruption across the broader economy arrive in the near-term, medium-term, long-term, or never?
15/n

Most AI maximalists who are well informed on AI think the world will look very different in 2-3 years than it does today. Like very, very different. They believe economy might be unrecognizable in 5 years and that policymakers must start preparing today for the disruption.
16/n

Those outside of tech see the many frictions and constraints on adoption in other industries and come to a different conclusion. They're more likely to see AI as a sustaining technology, improving productivity but not fundamentally changing the nature of most professions.
17/n

They point to regulation, liability, inertia, customer preferences, physical constraints, and legacy systems as barriers that slow adoption in the real world. And they point out that most jobs are not simply a collection of repeatable tasks but are something more complex.
18/n

I don’t know where to come out on it. AI researchers have access to models I don’t and a much greater appreciation for the trajectory of future models. But, many also have less exposure to the physical, regulatory and institutional constraints that shape much of the economy.
19/n

A host of policy responses have been proposed: UBI, wage insurance, taxing AI, workforce retraining, shorter workweeks, expanded social insurance, giving people equity in AI or broader stock market. But it’s hard to know what to do when we don’t know how this plays out.
20/n

The software world crossed an inflection point after models surpassed a certain threshold. Whether the rest of the economy follows remains an open question. It's going to be a fascinating few years to watch. Hopefully, they'll be good years as well.
21/21