AI agents are already commanding 75% to 100% of a human equivalent salary in labor-short markets, ahead of the timeline Tomasz Tunguz predicted in early 2025. Goldman Sachs data backs the structural shift: low-labor-cost stocks beat high-labor-cost stocks by 8 percentage points in 2025, and labor's share of GDP hit a record low of 53.8% in Q3 2025. Across the S&P 500, labor runs about 12% of revenues while software sits at 1 to 3%. That ratio is inverting.
The first-order story is output. The more important story is the cost stack underneath. Agents carry no FICA, no state unemployment insurance, no benefits, cutting 25 to 30% off the equivalent human cost before you factor in the Section 179 deduction, which covers agent software expenses up to $2.56 million. Training is faster because all materials can be delivered simultaneously. Capacity scales with inference spend, not headcount approvals. The piece also draws a clean line between full agent replacement and augmentation, where the sale targets the marginal hire rather than the whole team.
What makes this worth reading in full is the adoption pattern Tunguz describes: usage surges faster than either vendor or buyer expects, forcing a strategic pause for organizational redesign. That pause is where the real decisions happen. And because vendors can price at par to a human salary, there is no race to the bottom, at least not yet.
[READ ORIGINAL →]