AI is cutting G&A headcount in half, and the production evidence is already in

Nikesh Arora, chief executive of Palo Alto Networks, offers a rule of thumb that is blunt enough to be worth quoting directly: in the next three years, companies will probably have half the people in general and administrative functions, in marketing, finance, and HR, because of how much process management those roles involve. That is not a prediction about automation at some indeterminate future point. It is a near-term operational forecast from someone running a large enterprise.

The forecast has company. Jason Calacanis states that 50 percent of entry-level knowledge worker jobs are going away within one to five years. Separately, Calacanis describes an acquire-and-compress playbook in which companies buy software businesses, remove 80 percent of employees, and operate the remainder as an AI-first workforce. Whether that particular model scales beyond the companies using it is an open question, but as a proof point that the math is already being tested, it is hard to dismiss.

What makes Arora’s three-year figure credible is not forecasting consensus. It is the evidence already in production. Fred Turner, chief executive of Curative, describes a credentialing process that once took two to three months and cost roughly $50 per credential. An in-house agent built on Claude now completes the same work in about 12 hours at a cost of around 20 cents. Curative also canceled its $600,000 annual Salesforce contract after an internally built customer relationship management system, assembled through what Turner calls vibe coding, outperformed it. A contract-review agent handles redlining and signing without routing to outside counsel. Turner’s assessment of where this leaves the company is direct: “Today the current gen models can do every back office task we have at Curative. It’s just a matter of deploying them.”

Today the current gen models can do every back office task we have at Curative. It's just a matter of deploying them. Fred Turner

That last phrase carries the weight. The constraint Turner identifies is not capability. It is deployment speed. For enterprises still treating AI as a future investment, the implication is that the capability gap they are waiting to close has already closed in the functions where process volume is highest.

The compression shows up at larger scale too. Harry Stebbings describes a weekly capital allocation process covering pricing decisions across thousands of global markets that ran to 15 hours of work. The same process now takes two hours. Stebbings goes further, arguing that if that business’s scope were held static, it could perform its current operations with fewer people within five years. He also describes his own company’s plan: giving each of his best engineers $100,000 of tokens and, in return, cutting the size of his development teams by 30 to 40 percent.

The effect extends beyond headcount reduction to functional consolidation. Sarah Guo, co-founder of Conviction, describes a portfolio company where the person leading marketing built what he calls an autonomous marketing department, collapsing a multi-person function into a single operator. Quinn Slack, who leads Sourcegraph, watches smaller software companies hire product managers and marketers and calls it the old way of building a business. His point is structural: the historical model in which 90 percent of a company’s people handle overhead, leaving 10 percent focused on product, no longer reflects what AI makes possible. Tom Verrilli notes that work requiring an Amazon senior individual contributor one to two weeks can now be done in a single thread.

The function-by-function picture that emerges is consistent. The roles most exposed are the ones built around process volume: intake, verification, coordination, reporting, routing. Those tasks are exactly what current-generation agents handle well, and the cost differential between human and agent execution is not marginal. A $50 credentialing task at 20 cents is not a productivity improvement. It is a different cost structure entirely. Whether companies absorb the savings, redeploy the labor, or simply reduce headcount will vary. But the premise that the number of people required to run G&A functions at their current scale will remain roughly constant is, at this point, difficult to sustain.