Nathan Wilmers, Regina Catipon, Maxim Massenkoff, and Peter McCrory, in an Anthropic research article:

How has AI restructured work and employment at Anthropic? As a company at the frontier of AI adoption, Anthropic could be a leading indicator of how AI will shift work organization, hiring, and labor demand across the economy.

Anthropic has been hiring for a smaller share of junior technical positions. This occurred as hours worked by AI agents relative to employees rapidly expanded. By August interactive AI agent hours outnumbered employees’ 6 to 1.

As AI agents diffused through the company, Anthropic began producing more code per person, closing more sales per rep, and processing more contracts. Following the same employee over time, output rose as they adopted new AI tools.

This acceleration makes human attention the bottleneck for agent output. Even among non-technical employees, reading all the output of their Claude agents would take half their work week. In response, employees turn to “meta-work,” of building visibility tools and monitoring agent pipelines.

Changes in work organization are reflected in shifting priorities among hiring managers, who increasingly emphasize autonomy and judgment, relative to technical domain expertise and implementation…

Anthropic has responded to these changes by setting up an early warning system to track jobs at risk of disruption, and as jobs evolve the company has prioritized internal redeployments. These actions, taken at the frontier of both AI development and workplace change, show how specific organizational commitments can shape the impact of AI on employment. But they also raise difficult questions about exactly what those commitments should be.

Anthropic has not “responded” to these changes as much as it has caused them itself. Not by developing Claude, but by intentionally reducing hiring as a result of greater emphasis on “autonomy and judgment.” Setting up an early warning system for jobs being intentionally disrupted is cyclical in some sinister way.

But the emphasis on “autonomy and judgment” is not unwarranted, nor inherently negative — it just doesn’t need to come at the cost of entry-level jobs, which are fundamental to a functioning economy. Junior-level interns and full-time employees can, too, exercise judgment and take advantage of autonomy, and giving them the opportunity to do so at one of the most influential technology companies of our generation is a great way to positively influence artificial intelligence’s economic impact. What I’m saying here is that if we never allow entry-level workers to hone and exercise their judgment, relegating them to lower-level jobs that are, by Anthropic’s own admission, fading, we’ll have a concerning deficit of senior-level skilled workers in a few decades.

There’s a lot to like about the current trends in tech hiring. “Technical” skills — or, more specifically, the credentials to prove those skills — have never been fairly distributed among laborers. There are many more talented computer scientists than those who come out of Stanford University, many of whom lack degrees altogether. Technical interviews, I’d argue, are moot in the age of automation. As AI reshapes hiring, I hope to see emphasis on university prestige and technical interviews fade into obscurity. But this does not mean that we shouldn’t be hiring entry-level employees, nor that there isn’t a way to measure their judgment and capabilities without formal credentials. If the emphasis is on autonomy and judgment, hire free thinkers and people with great ideas. Test them on those ideas, not how well they’ve memorized the syntax of C++. If the latter is what recruiting is optimized for, then yes, there’s no point in hiring an entry-level computer science major.

But computer science is not a vocational degree, and candidly the only people who think it is are venture capitalists who studied it for the money. To them, I say, Get Out of Technology. Computer science comprises a unique blend of the liberal arts, humanities, and engineering. There’s a reason computer science isn’t called “information technology,” and why most computer science courses hinge so strongly on theory, not application. Computer science students don’t “learn to code for four years,” as one ill-informed internet commenter I found said. Most of them — the ones who are least likely to drop out — have already learned and built software by themselves. To gain admission to a relatively prestigious university’s undergraduate computer science program, it is almost a prerequisite that you’ve already become a software engineer by the time you’re 16.

It’s time to start treating computer scientists like philosophers and drop the act of pretending computer science was ever about software engineering. When autonomy and judgment are paramount to getting a job at one of the most influential AI companies, computer science students are the most natural fit. They learn over four years to not only engineer software, but think deeply about computation and automation itself. Good computer scientists know not just how to automate, but what should be automated at all. Learning programming languages and solving hundreds of LeetCode problems — as the research paper posits — does not teach this. Test computer scientists on the judgment they should have spent four years learning about and refining. New computer scientists are not entry-level software engineers, and when we stop treating them as such — or expecting them to assume the title of software engineer — suddenly their education carries much more value.

We should absolutely be hiring entry-level computer science majors to think through the toughest questions about the philosophy of computing, automation, and data. We should be using their four years of experience to shape AI development, not to engineer software. And once the industry catches up to this — I think it will eventually — it’ll prove that there has never been a better time to study computer science than during an era where computers are poised to usher in a new societal order.