A March 2026 essay by technologist Rajiv Pant poses a pointed question about the agentic AI era: when workers achieve 3x, 5x, or 10x productivity multipliers through AI tools, who actually captures that surplus value? Pant, currently President of Flatiron Software and Snapshot AI, grounds the argument in substantial empirical evidence — McKinsey's estimate of $2.6 to $4.4 trillion in annual AI-driven economic value, PwC's 2025 Global AI Jobs Barometer showing productivity growth nearly quadrupling in AI-exposed industries, and his own documented case where two AI-augmented engineers delivered the output of a full traditional team at roughly 3.6x productivity. His thesis: the default outcome, absent deliberate employer action, is that gains flow entirely to capital.
The essay's sharpest contribution is the concept of "synthesis engineering" — the human skill of knowing when to trust AI, when to override it, and how to frame problems for it effectively. Pant draws on the BCG "Jagged Frontier" study, which tracked 758 consultants using GPT-4 at Harvard Business School, to show that this skill is both real and decisive. The same tool, the same people, on the same day produced a 40% quality improvement when used within AI's capability frontier and a 19% quality degradation outside it — a 59-percentage-point swing determined entirely by human judgment. Garry Kasparov's centaur chess insight, that a weaker human with better process outperforms a stronger human with worse process when augmented by a machine, reinforces the point. PwC's data adds market validation: AI-skilled workers commanded a 56% wage premium in 2025, double the prior year's figure.
Pant engages the strongest counterargument directly: employers paid for the AI infrastructure, bore adoption risk including the productivity J-curve documented by MIT and NBER researchers, and financed integration and governance costs. EY's 2025 survey found 96% of AI-investing organizations seeing gains, most of which are being reinvested into further AI rather than shared with workers. <a href="/news/2026-03-14-meta-weighs-20-workforce-layoffs-to-offset-rising-ai-infrastructure-costs">Some organizations are pursuing a third path: using AI gains to justify layoffs rather than reinvestment</a>. But Pant argues this logic proves too much — by the same reasoning, employees would deserve no credit for productivity gains from computers or spreadsheet software, a position no serious economist defends. He situates the current moment within a decades-long structural divergence documented by the Economic Policy Institute: from 1979 to 2019, U.S. productivity grew 59.7% while typical worker compensation grew only 15.8%.
There's a tension the essay doesn't resolve: the 3.6x productivity case study Pant cites as evidence that workers deserve a share of AI gains comes from his own firm, where he is the employer of record. <a href="/news/2026-03-14-polsia-ai-solo-founder-3-5m-arr">Some startups are taking the approach even further, claiming to generate millions in revenue with AI agents and no human workers at all</a>. The essay doesn't disclose what those two engineers earned relative to the traditional team whose output they replicated — the most important empirical question in the piece, left unanswered. Still, the closing argument is practical and pointed: employers who fail to share AI productivity gains risk burning out the high-judgment workers whose synthesis engineering skills make those gains possible, ultimately eroding the productivity improvements they're trying to capture.