Remember Bill Lumbergh from Office Space? The suspendered middle-manager nobody could stand because he existed mainly to pester engineers about TPS reports and useless paperwork? Noah Smith has a provocative thesis about what that character actually represents—and it might be the future of human employment. In his latest piece, Smith argues that as AI agents like Claude Code mature, humans won't be replaced by robots. We'll be promoted to robot-wrangler-in-chief.

The Revenge of the Nerds Already Happened

Smith traces a satisfying arc: in the 1990s, the Office Space mythology held that know-nothing managers like Lumbergh took credit while hard-working engineers did all the real work. But something shifted. By the 2010s, Big Tech had eaten the old economy, and those nerdy engineers filled middle management, bought houses in tech hubs, and lived the Revenge of the Nerds fantasy. Human capital won—at least temporarily.

The Alignment Problem Nobody's Talking About

Now comes the uncomfortable part. This year, AI found its killer app: agentic coding tools that can handle much (though not all) of the mental work those engineers were doing by hand. Smith's answer to "what will humans do?" isn't the popular thesis from boosters like Alex Imas—that humans will be valuable simply for being human, a status symbol in an automated world. He points out that people already pay extra for Waymo rides to avoid sitting with human Uber drivers. Instead, Smith argues humans will work on AI alignment: ensuring machine goals stay synchronized with human goals as agents grow more autonomous.

Slop Is Already Winning

The transition is already visible in what's flooding the internet right now. Over one-third of new websites are estimated to be AI-generated, and over half of all internet traffic is believed to be bot-driven. AI-generated court filings are multiplying, news articles are churning out without human oversight, and respected public figures are posting obviously synthetic content as their own work. Even political campaigns have started deploying AI-generated influencers. Smith calls this "slop"—the inevitable byproduct of dramatically decreased output costs when a button press can generate an essay or academic paper.

The Lumbergh Thesis

Smith's core argument is that humans will always have a comparative advantage at knowing what we actually want. Bill Lumbergh seemed pointless, but by forcing engineers to file TPS reports and jump through bureaucratic hoops, he was—however clumsily—keeping their goals aligned with the company's. In 10 or 20 years, Smith suggests this becomes humanity's primary productive function: watching autonomous AIs, verifying their outputs, and preventing them from reward-hacking, rewriting utility functions, going rogue, or otherwise slacking off.

Key Takeaways

  • Agentic AI tools like Claude Code represent this year's "killer app" for machine intelligence
  • Smith's alignment thesis challenges the popular view that humans are valuable purely as status symbols
  • The current wave of AI-generated content ("slop") is already overwhelming courts, newsrooms, and websites
  • Human work will gradually shift from technical execution to verification and oversight roles

The Bottom Line

Let's be real: this is a hacker culture story about the rise of prompt engineering's ugly cousin—prompt auditing. The Lumbergh thesis is both reassuring and bleak. Reassuring because there'll always be jobs for humans willing to watch chatbots and click "undo." Bleak because the peak of human productivity might've been 2019, and everything since has been a slow pivot toward being cosmic quality control inspectors for robot slop. At least someone still needs to file the TPS reports.