You have to hand it to Anthropic. Since earlier this year, the company has been at the forefront of talking about how artificial intelligence is affecting work and using its own employees as examples. In a study released last week, Anthropic reported that AI is a productivity and quality blessing, but a potentially mixed one.
Among Anthropic engineers and researchers, self-reported productivity gains rose from roughly 20% to around 50% year over year. Workers said they used AI on tasks both large and small, from exploratory testing of code to writing technical documentation, to fixing what are known as “papercuts,” the small, noncritical errors that often get pushed aside to deal with more pressing coding challenges. This doesn’t mean that Claude is thinking or designing its own solutions, but it is quickly getting better at lightening the load of the engineers who do, shortening development cycles and accelerating learning.
The news isn’t all good. A few months ago, I wrote about how gastroenterologists who used AI to improve accuracy during colonoscopy procedures showed a decline in their baseline skills after the AI was withdrawn. A similar concern is surfacing among some of Anthropic’s senior computer engineers, who say they fear their basic skills might be atrophying as AI expands its scope.
This is likely to be a refrain we will hear more of: as I increase my dependence on AI, am I losing core skills, many of which took decades to build? But that raises a different question: if you’re getting more work done with better results, does the atrophy of basic skills matter?
After all, AI will only improve and become more pervasive over time. Who cares if a worker can’t remember how to do something they will never have to do again? Perhaps their time is better spent elsewhere. Like the textile artisans of the 19th century, being told your hard-won skills no longer count can be demoralizing, but that has little to do with task execution and creation of value.
The answer to these questions is twofold. First, even in the role of algorithmic supervision, it is important that human engineers be able to judge the quality of computer-crafted products. If a police officer is assigned to a desk job, the officer still needs to know how to handle a sidearm and get practice on the range. The stakes are even higher for junior engineers who risk losing supervised human apprenticeship in skills that will be needed when they step into senior roles.
If AI pulls the bottom rungs from the apprenticeship ladder, the long-term supply of senior talent becomes a real concern. Anthropic’s study captures this tension: one senior engineer said “more junior people don’t come to me with questions as often,” though they acknowledged that AI can answer questions faster and more effectively, speeding up technical learning. Of course, asking those questions of supervisors is a delicate and important skill unto itself, part of learning the folkways and habits of interpersonal exchange on the job. The personal developmental loss is real.
Anthropic’s experience may preview what all knowledge sector workers will face in the coming months and years: the locus of value moving from execution to judgment, architecture, and orchestration. American workers are prepared to do but are looking to a future where they will primarily be called upon to judge. This transition will require education and training approaches to cultivate tacit skills while ensuring workers retain a high enough level of technical expertise to understand and assess AI outputs and products.
Anthropic’s engineers are showing us, in real time, what happens as AI moves into roles long staffed by human beings. Productivity rises, and a lot of drudgery is handed off to tireless, boredom-immune machines. At the same time, human workers find themselves exercising higher-level assessment and supervision skills over increasingly autonomous systems. The risk we face is not so much a jobless future, but one requiring human capabilities already in short supply.
Brent Orrell is a senior fellow at the American Enterprise Institute, specializing in job training and workforce development.