Key Takeaways

  • Naveen Rao projects the AI inference market will reach $10 trillion within a decade as dropping token costs trigger Jevons Paradox across the global economy.
  • Cutting token delivery costs by an order of magnitude expands addressable use cases by more than 10x, turning single-trillion sectors into multi-trillion markets.
  • Rao predicts robotics will be 100 times larger than coding agents because physical automation captures value from the broader labor market rather than desk work alone.
  • Harry Stebbings challenges near-term robotics timelines, while Rao expects working prototypes for specific physical tasks in 2 to 3 years and commercial markets in 5 to 7 years.

The Disagreement

Stebbings and Rao collide on where founders and investors should direct capital today: digital workflow automation or physical robotics.

Stebbings remains skeptical about near-term robotics timelines. He views humanoid robots as capital sinks with distant commercial payoffs, preferring the safety of proven software workflows like AI coding assistants.

Rao rejects that caution. He argues that traditional pure-software software-as-a-service faces severe threats, while the real prize sits in physical work: “I actually think robotics is going to be 100 times larger than a coding.” When Stebbings questions whether the technology can ship fast enough to justify venture math, Rao sets an aggressive schedule. He expects “working prototypes for some of these tasks in the next 2 to 3 years and the market will emerge in 5 to seven years.”

The tension centers on market scale versus execution risk. Stebbings worries about the cash required to build hardware that might take a decade to function reliably. Rao sees manual work as the largest pool of economic value on Earth, which makes the entry valuation secondary: “Guys when we're talking about trillion plus dollar TAMs the price is almost irrelevant.”

Who's Right (and When They're Wrong)

Rao is right about the terminal economics, but Stebbings is right about early-stage survival.

Rao's thesis relies on Jevons Paradox. As inference hardware improves and cost-per-token falls, demand will explode instead of shrinking. In Rao's words, “the reason that I think it's a fantastic market to be in is that if we knock off a zero, one order of magnitude efficiency on delivery of a token, that effectively 10xes the market in some ways. So, something that was a trillion dollar market, now all of a sudden you have way more applicability. The base of potential uses just increases more than 10x.” Applying that dynamic to physical labor creates an unprecedented market. Coding is a tiny fraction of global payroll; moving boxes, assembling components, and maintaining physical infrastructure represent tens of trillions of dollars.

Yet Rao's timeline ignores the graveyard of hardware companies that ran out of cash waiting for software to catch up. Building physical systems requires specialized hardware co-design, custom actuators, and high safety margins. If you build for a five-to-seven-year commercial market, your seed funding will vanish long before customers deploy units at scale unless you raise hundreds of millions.

Stebbings is right for founders who cannot raise mega-rounds: coding agents produce recurring software revenue today. But Rao is right about where defensible enterprise value will sit. Software that merely assists programmers is a feature that large foundation model providers can absorb. Replacing physical human labor creates physical moats that pure software cannot duplicate.

What to Do With This

Stop building thin wrappers around coding models and audit physical bottlenecks in non-tech industries this week. Pick one manual, repeated physical workflow in warehousing, food preparation, or field maintenance where human labor costs exceed $50,000 per worker annually. Interview three operations managers to determine the exact failure tolerance and unit economics required to deploy a single-purpose robotic task cell before touching any CAD software.