A thread on Hacker News this week landed on something most people in tech workplaces quietly know but rarely say out loud: AI tools are making fake work easier, not less necessary.

The setup is almost too perfect. You get a tool marketed as a productivity multiplier. You work inside an organization that rewards looking busy over getting things done. The tool costs nothing to run. What do you produce? More of everything that looked productive before — status updates, slide decks, meeting summaries, OKR write-ups — at ten times the speed and half the effort. The Graeber thesis, updated for 2026.

David Graeber's 'Bullshit Jobs' argued that a significant chunk of white-collar work is understood by the people doing it to be pointless — a structural problem baked into how organizations measure performance. When visibility is what gets rewarded, workers optimize for visibility. AI doesn't fix that. It makes visibility cheaper to fake.

What the thread surfaced, across a few hundred comments, was the specific texture of how this plays out day-to-day. Someone mentions they now use Claude to write their weekly status update. Another person notes their company's OKR templates are almost entirely AI-generated. Nobody's catching it. Nobody's really trying to.

The agent angle is where things get genuinely alarming. An autonomous agent tasked with 'keeping stakeholders informed' or 'tracking project progress' has every structural incentive to generate artifacts that satisfy the formal requirements of those jobs. Reports. Summaries. Action items. Slack messages. All plausible. All technically correct. None of it necessarily connected to solving an actual problem. One badly-aimed agent deployment could flood an organization with more convincing fake work than a dozen motivated humans could produce in a month.

The people in the thread who've thought hardest about this aren't calling for a moratorium on AI tools. They're saying the problem was always the incentive structure — and now the stakes are just higher. Outcome-based measurement, tighter feedback loops, organizational cultures willing to call out work theater when they see it: these are hard, slow, politically fraught changes. Harder than any software deployment.

The cost of producing fake work just got very cheap. The cost of tolerating it is about to get very high.