How AI agents could destroy the economy

On Sunday, a research outfit called Citrini Research published a striking analysis outlining how agentic AI could cause widespread economic damage within the next two years. The piece is framed as a document written from a near-future vantage point — a report from two years ahead — describing a world where unemployment has doubled, and the stock market’s total value has dropped by more than a third. In Citrini’s imagined account, the core dynamic is described like this:

AI capabilities improved; companies needed fewer workers; white-collar layoffs increased; displaced workers spent less; margin pressure pushed firms to invest more in AI; AI capabilities improved…
It was a negative feedback loop with no natural brake…The system turned out to be one long daisy chain of correlated bets on white-collar productivity growth.

It’s a different flavour of bear case than the usual AI doom narratives. Rather than focusing on Skynet-style “misalignment,” Citrini’s scenario centres on something slower and arguably more mundane: the economy gradually coming undone as AI agents are embedded across businesses and markets. The analysis zeroes in on what happens when AI agents become widely adopted as operational infrastructure — and in particular, what it would mean once companies begin replacing outside contractors with cheaper in-house agents.

The concept overlaps with the “Death of SaaS” argument, in which AI agents reduce demand for traditional software by handling tasks directly. But Citrini stretches the thread further, arguing that the risk extends beyond SaaS into any business model built on optimising transactions between companies — essentially, firms that make money by sitting between organisations and smoothing or coordinating business processes.

Unsurprisingly, the write-up has sparked significant debate online. Plenty of people aren’t convinced — Citrini itself presents the piece more as a structured scenario than a confident forecast — but it’s also surprisingly hard to identify the exact point where the chain of logic definitively breaks. The feedback loop is coherent enough that objections often end up feeling like “I don’t think it would happen that fast,” rather than “this couldn’t happen at all.”

On a personal level, I’m sceptical that most companies are ready to fully delegate purchasing and procurement decisions to AI agents, regardless of how capable the systems become. But Citrini’s framing complicates that objection: in the scenario, many of the most affected decisions are already being outsourced to third-party contractors. If those decisions are already being handed off to external firms, the jump to handing them off to lower-cost internal AI agents may be less far-fetched than it initially sounds.