Be Cautious of Digital Deskilling

      Home » Blog » Caution Against Digital Deskilling

      Last week, Boris Cherny, the creator and lead of Anthropic's well-known Claude Code programming agent, shared a thread on X about his personal use of the AI tool in his work. This sparked significant conversation. "What started as a casual insight into his terminal setup has turned into a viral declaration on the future of software development," noted a VentureBeat article regarding the situation.

      According to Cherny, he operates five instances of the coding agent simultaneously, each in its own tab on his terminal: "While one agent executes a test suite, another refines a legacy module, and a third composes documentation." He swiftly navigates through these tabs, providing additional guidance or gentle nudges to each agent as required, reviewing their work, and asking them to enhance their output.

      One commenter on the thread likened the process to playing the famously fast-paced video game Starcraft. The VentureBeat article characterized Cherny's operation as akin to that of a "fleet commander." It all appeared to be an enjoyable experience.

      However, as a software developer, I would remain cautious of such demonstrations.

      In his 1974 book, Labor and Monopoly Capital, influential Marxist political economist Harry Braverman argued that the expanding "science-technical revolution" was being leveraged by corporations to increasingly "deskill" workers, leaving them in "ignorance, incapacity, and thus in readiness for machine servitude." The more proficient tasks are outsourced to machines, the more manageable the employees become.

      It's difficult not to perceive echoes of Braverman's deskilling argument in something like Cherny's AI programming demonstration. A scenario where software development is reduced to mere management of enthusiastic yet chaotic digital agents constitutes a reality in which a once vital economic sector is diminished to fewer, lower-paying jobs, as managing agents requires significantly less expertise than creating sophisticated code from scratch. Consumers wouldn't benefit either, as the resulting software would likely be less reliable and innovation would stall.

      The only group that would clearly gain from developers' deskilling would be the tech companies themselves, as they could reduce one of their largest expenses: their workforce.

      Boris Cherny serves as a senior technical lead at Anthropic, managing a sizable team and likely possessing a considerable amount of stock options in the company. Naturally, he’s enthusiastic about agents taking over programming tasks, but that doesn’t mean we need to share his excitement.

      —

      P.S. I don’t intend to dismiss the benefits of AI tools for programmers. I’ve spoken with many developers who have found significant value in using AI to aid (reportedly) in expediting programming tasks. What raises my suspicion is the assertion that transitioning to a system where one merely delegates tasks to agents is somehow the logical next step in programming productivity. It may appear appealing in the present, but something deeper and more concerning could be hiding beneath these impressive demonstrations.

Other articles