October 20, 2025
Work today is human-shaped. Over the last 600 years, we've built an entire economy optimized for how humans operate. This infrastructure is designed to amplify a single human’s efforts. A capable person inside an org can drive far more change than they could otherwise alone, not because they've suddenly become more talented or capable, but because they're plugged into machinery designed to multiply their impact. While he was doing similar things before, it was when Clive leaned into and bootstrapped from East India Company’s resources that he toppled over an entire regime. Hardy gave Ramanujan the infrastructure that helped him produce so many theorems and formulas.
And while in isolation it can be hard to drive change, we know that humans are capable of bootstrapping from very low baselines, I don’t think we can say the same for LLMs yet. Even now they require humans to build massive infrastructure for them to produce even little economic value.
Markets become efficient when making them efficient is profitable. LLMs are barely a few years old. They don’t yet have an infrastructure that plays to their strengths. You can have a ton of llm agents dropped into human shaped holes, but just having a lot of them doesn’t create a lot of value. Abundance doesn’t mean autopoiesis.
Over time, we will build better scaffolding, better infrastructure for agents. Then, at some point, the efficient shape of work will drift away from human-centric clusters to being reorganized around agents (for example, something that leverages ability to work 24*7, have massively parallel decision trees, instant context loading, high utilization, no need for a lot of social coordination infrastructure, etc). In that world, when the primary infrastructure is designed for the strengths of the agents, it’ll be human capabilities that’ll look spiky.
I think this process will be somewhat slower than what many people expect, because of path dependence and switching cost. I wonder if things played out this way during the industrial revolution. Did the first factories simply gather craftsmen under one roof, replicating workshop methods at scale? How long did it take to discover that machines shouldn't mimic human motions but could operate on entirely different principles?