Damon Binder, a senior researcher at the philanthropic research group Coefficient Giving, has published an argument that artificial general intelligence would not stay confined to knowledge work. His claim, laid out in a memo for AI Frontiers: whatever lets a system hold down every remote job (steering a process moment to moment, reasoning about where things sit in space, anticipating how objects behave, and learning past the end of training) is the same repertoire a robot needs on a factory floor. If that premise holds, Binder estimates a fully automated economy could double its physical output roughly once a year.
The stakes are bigger than a robotics forecast. Binder is arguing that automating cognitive labor removes the one constraint that has always capped physical production: a fixed human workforce. Machines that can build more machines, including the robots that replace workers, sever growth from population.
The reasoning runs in a clean chain. Start with a definition of AGI as a system you can hand a task the way you’d hand it to a competent contractor, at a cost that competes with a human hire. Binder then argues that a system able to do that across remote work already possesses the perception and control skills robotics needs, so building the actuators to pair with that intelligence is comparatively straightforward. From there he leans on input-output analysis, the Commerce Department’s accounting of what each of 402 industries buys from every other, combined with a growth-rate method developed by mathematician John von Neumann, to argue that once workers can be manufactured like any other capital good, output compounds fast: doubling in well under two years even after building in construction delays, or every twenty months if half of everything produced goes to human consumption rather than reinvestment.
The load-bearing claim in this argument, and the one Binder does the least to prove, is that hardware is not the bottleneck. His evidence is historical rather than demonstrated at scale: industrial arms have exceeded human precision for half a century, and remote-controlled machinery has handled radioactive material since the 1940s, built underwater structures, and performed surgery, whenever a human operator could direct it. That shows competent control unlocks skilled physical work. It does not show that humanoid robots can be manufactured at automotive volumes today. Binder’s cost case rests on Unitree’s roughly $13,500 G1 humanoid and Tesla’s reported $20,000 target for Optimus, prices that assume a production ramp neither company has hit yet.
This is exactly where the obvious counter-pressure sits: current robotics manufacturing throughput. Binder’s own transition model partially concedes the point. It acknowledges that “the machine tools needed to build more are themselves in short supply” and puts robot production at roughly nothing for years before it reaches the tens of millions, with electricity generation, a precondition for running that fleet, taking about four years to double once and two more to double again. Where the argument falls short is in how casually it treats the timeline for closing that supply gap: its scenarios assume new industrial capacity comes online in zero, six, or twelve months, a range shorter than the multi-year lead times real chip fabs and heavy-equipment plants currently require. The model absorbs the throughput problem as a scenario input rather than as evidence that the buildout itself might take longer than a year or two to even begin compounding.
Binder is explicit that he assumes no new science, no recursive self-improvement, and 2017-era American production methods, all of which he calls conservative rather than aggressive assumptions. That framing matters for how operators should read the piece: it is a capacity argument, not a timeline prediction for when AGI arrives. Teams evaluating robotics or humanoid hardware bets over the next year should treat manufacturing lead times, not model capability, as the variable to watch, since Binder’s own numbers suggest that is where a real industrial transition would actually stall first.
Damon Binder’s essay was published by AI Frontiers on August 11, 2026.
