Last week, Boris Cherny, the creator and head of Anthropic’s popular Claude Code programming agent, posted a thread on X about how he personally used the AI tool in his own work. It created a stir. “What began as a casual sharing of his personal terminal setup has spiraled into a viral manifesto on the future of software development,” explained a VentureBeat article about the incident.
As Cherny explained, he runs five different instances of the coding agent at the same time, each in its own tab in his terminal: ‘While one agent runs a test suite, another refactors a legacy module, and a third drafts documentation.’ He cycles rapidly through these tabs, providing further instruction or gentle prods to each agent as needed, checking their work, and sending them back to improve their output.
One user, responding to the thread, described the approach like playing the famously fast-paced video game Starcraft. The VentureBeat article described Cherny as operating like a “fleet commander.” It all seemed like a lot of fun.
But here’s the thing: If I were a software developer, I would be wary of any such demonstration.
In his 1974 book, Labor and Monopoly Capital, the influential Marxist political economist Harry Braverman argued that the expanding “science-technical revolution” was being exploited by companies to increasingly “deskill” workers; to leave them in “ignorance, incapacity, and thus in fitness for machine servitude.” The more employees outsource skilled activity to machines, the more controllable they become.
It’s hard not to hear echoes of Braverman’s deskilling argument in something like Cherny’s AI programming demo. A world in which software development is reduced to the ersatz management of energetic but messy digital agents is a world in which a once important economic sector is stripped down to fewer, more poorly paid jobs, as wrangling agents requires much less skill than producing elegant code from scratch. The consumer would fare no better, as the resulting software would be less stable and innovation would slow.
The only group that would unambiguously benefit from deskilling developers would be the technology companies themselves, which could minimize one of their biggest expenses: their employees.
Boris Cherny is a senior technical lead at Anthropic who manages a large team and likely owns a significant amount of stock options in the company. Of course, he’s excited about the idea of agents replacing programmers, but that doesn’t mean we have to share his enthusiasm.
—
P.S., I don’t mean to deny the value of AI tools for programmers. I’ve talked to many developers who have found great utility in using AI to help (apparently) speed up programming tasks. What makes me suspicious is the claim that shifting to a world in which you just assign agents work is somehow just the natural next step in programming productivity. It might seem cool in the moment, but something more profound and dark might be lurking beneath these gee-whiz demos.
I also would be extremely wary about this. AI can be helpful in a lot of ways, for sure, but it seems like happily training your own replacement, or turning everyone into mini-managers of robots that produce work of questionable quality. And yes, ultimately the end user who benefits the most is the people doing the hiring, so naturally I’m a little suspicious about the AI push. Let’s all be sensible when using these tools, stay critically thinking, and follow where the money goes.
I think there are definitely concerns here, but the pay for those who can successfully wrangle agents should not be one of them. This is turning the 10x developer into the 100x developer and while I doubt they’ll get a 10x salary boost, the engineers with the taste, judgement and skill to run tens of agents to efficiently deliver robust, production quality code are likely to see a material bump in comp as they’ll be in extremely high demand for the value they can deliver.
Anyone who thinks it is (or will be) easy to thoughtfully and rigorously specify requirements for an application in a way that is maintainable, scalable and impactful should try it for a couple of days!
AI companies are shifting from trying to replace workers to convincing them that they can’t do their job without them. “AI won’t take your job, it’s people with AI who will take your job.” It feels like more of a psyop to force dependence on yet another subscription service.
How is this trend different from what happened with blacksmiths and the evolution of metal working with the development of cutting torches, power hammers, laser cutters, CNC machines…? In the early stages of those modern tools, they were far from perfect, but they evolved and continually improved. With those advancements, it opened the doors to growth in new areas beyond just working with metal, and people were still needed to fill the roles that were created as a result.
I’d say that AI is replacing thinking itself. The industrial revolution covered a lot of the need of physical labor and helped us advance beyond it. Now that both human cognitive skills and human physical skills can be replaced, I find it difficult to imagine exactly what we are going to do.
I’m not too sure, however, what the call to action is. What if it is the fact that software development is going in the direction of managing multiple sub-agents? What if one’s employer insists that the developer manage multiple sub-agents? How does one hold on to their job without being forced to “de-skill” themselves in this world of sub-agent management?
You could bring 14,000 other quotes warning against automation besides the one from a Marxist.
As a linguist, I have been observing with a lot of concern the replacement of translators by ai-driven platforms, and the question of what should be taught in a foreign language curriculum is serious matter.
So, I am sincerely curious to see whether computer programming itself may be affected the same way.
I agree with you Cal on the thesis that as a whole, for Software Engineers in the world, reliance on these agents will lead to worse outcomes for people, good outcomes for capital and corporate leaders. But the issue is that on an individual level, the agents do accelerate productivity – they give you a leg up on your peers, so you have no choice but to use them. So short of total global coordinated action that is unlikely to occur, the “greedy algorithm” followed by every SWE in the world in terms of productivity seems like it will inevitably lead to the deskilling outcome.
Perhaps it has lead to it already.