Anthropic reports reducing recruitment of entry-level engineers

The Anthropic Institute’s newest report says that the company’s junior recruitment has begun to slow down.

Anthropic has been hiring for a smaller share of junior technical positions.

I’m unsurprised. Anyone who has tinkered with coding agents over the last 10 months has seen this coming. But my read of many top-tier economists’ reactions is that they were largely dismissive of this concern. My suspicion for their dismissiveness is that they never spent time with coding agents in this period despite understanding that frontier models were increasingly capable. Heck, I know STEM academics who remain dismissive about AI capabilities. Frontier labs’ revenues have grown so rapidly that their valuations match some of the world’s largest companies. Again, if you have been playing with coding agents, you will know that this is not some outlier event with no clear explanation. So, I have heard the death knell for entry-level jobs quite loudly. For this reason, I have been repeating my concerns to my undergraduate students that they need to prepare differently for a likely upheaval.

Anthropic has expanded rapidly. But it has hired less per dollar of revenue than an earlier generation of large tech companies. At five years old, Anthropic takes in between 2–4 times as much revenue per employee as Meta and Google as they crested in their own rapid growth.

Junior-level recruitment at Anthropic might be dropping because they need fewer people. Instead, employees do more than what they were initially recruited to do; I read this as non-technical Ant employees doing some technical work as they train into becoming Claude managers:

 From sales to legal, human employees are expected to do more and different tasks. Roles broaden to include building products, managing agents and tackling the strategic and interpersonal work that remain beyond the scope of the best LLMs.

There are many reasons to be worried if you’re a student in higher education, especially if you’re in STEM. I suspect those training in domains affecting the physical world (civil, mechanical, aerospace, manufacturing) could be offered some respite, others have bleaker predictions of an industrial explosion — i.e., full physical labour automation — within a decade. While I remain quite skeptical of such automation, I will concede my skepticism could be misplaced. Why? If you’d asked me precisely a year ago if we’d have the super-capable coding agents that we now have, I’d have laughed you out of the room.

Of course, on the very first day of this year, I changed my mind after spending 36 hours with coding agents beginning around 11 am on New Year’s Eve. That is how shaken and excited I was by them.

There is more grim commentary in the report:

Reading these patterns forward and out across the economy suggests that AI diffusion could risk worker displacement.

I think this will hit academia pretty hard and pretty soon. My sense is that most academics and staff I have interacted with remain unaware and therefore dismissive about current AI capabilities. The majority do not: have subscriptions giving them access to top-tier models in their chat apps; work with coding agents; or keep up with the latest developments on Twitter — I will concede that each of these feels like a full-time job — which is why they hold their current view. The higher-ed system is unprepared for AI’s impacts across the sector as it diffuses through the economy, if (or, more likely, when) it happens.

One consequence of staying on this course of unpreparedness is that academic jobs will start to vanish as student applications drop in reaction to a shrinking entry-level job market. Here, I anticipate computational STEM will be hit first as, like software engineering and computer science, these jobs will increasingly be automated. Soon, whole departments could fold and, eventually, entire universities.

This all sounds grim, but doesn’t have to be so. Anthropic is already preparing for this future:

Anthropic is standing up an ongoing early warning system to monitor jobs facing AI-driven transformation.

Universities should follow suit (which needs them to become aware!). If they do so, they will realise that this is also insufficient to safely circumnavigate longer-term issues:

None of these actions will prevent powerful AI from disrupting jobs.

Preparedness will buy universities time to course-correct towards a different model where students graduate with agency, not just a certificate. Specifically, I think universities ought to pivot towards becoming incubators of talent, where students work on creating single-person (or few-person) startups with agents at the core; these startups could eventually realise where there are person-shaped holes which will result in creating meaningful and niche human jobs.

At the start of this year, I warned my students that entry-level jobs will shrink, sensing the seismic shift that coding agents were bringing about. I also advised they seriously think about expanding the scope of their final year projects with coding agents into something that could outlast their degree and provide them a pathway to secure self-employment. This hasn’t borne fruit yet but I am seeing evidence of change in this year’s cohort of final-year students in my class.