I run about a dozen services off a used office PC: an IoT sensor network, trading bots, a satellite-imagery report engine, a peer-support site for male abuse survivors, a farm-monitoring tool, an app store. Ten years ago that was a small team’s job. Today it’s one person and a robot. That gap is worth measuring, because it’s the same gap that reordered human civilization twice before.
The software delta: 2015 to 2026
Start with the honest baseline. In 2015, a solo developer shipping something non-trivial spent roughly a week hand-coding a feature — reading docs, hunting Stack Overflow, writing boilerplate, debugging. Maybe 50 to 150 lines of genuinely useful code on a good day, once you subtract the dross. A small SaaS MVP was a two-to-six-month project.
In 2026, with a capable AI agent at the keyboard, that same feature is hours — sometimes minutes. Boilerplate is free. Docs-reading collapses into the tool. The MVP is days to a couple of weeks.
But let’s be honest about the multiplier, because the “100x” hype is a lie. For greenfield, well-trodden work it’s genuinely 10–30x. For novel, hard, or legacy-tangled work it’s more like 2–5x, because the bottleneck moves off your fingers and onto your judgment — integration, verification, knowing what to ask for. Those still sit squarely on the human.
Call it ~15x effective output for an indie who knows what to ask for. The real shift isn’t speed — it’s that the cost of attempting an idea dropped from “burn a month of weekends” to “try it tonight.”
We’ve pulled this lever twice before
Agriculture, ~10,000 BC
The energy surplus per farmer rose enough that not everyone had to grow food. That surplus is civilization — it bought specialists: scribes, priests, soldiers, artisans. The lever wasn’t “more grain.” It was freeing human attention from subsistence.
The Industrial Revolution, ~1760–1840
Output per textile worker jumped 50 to 100x with mechanization. But here’s the part everyone forgets: it took two or three generations for wages and living standards to actually rise. Economists call it the Engels’ pause — productivity soared for decades before the median worker saw a cent of it. Displacement came first. Broad benefit came much, much later.
The same script, three times
| Revolution | What got automated | Multiplier | Who captured value first | Lag to broad benefit |
|---|---|---|---|---|
| Agriculture | Caloric subsistence | surplus → specialists | Landowners | Centuries |
| Industrial | Muscle / mechanical labor | ~50–100x | Capital owners | ~2–3 generations |
| AI software | Cognitive / symbolic labor | ~10–30x (task-dependent) | Platform owners + early adopters | Unknown — compressing fast |
The pattern that matters: each revolution automated a category of human effort, and the displaced flowed into newly-cheap adjacent work. Farmhands became factory workers became service and knowledge workers became… what?
The humanitarian read
The optimistic case is the agriculture parallel. If cognitive labor gets cheap, the surplus buys a new class of specialists — millions of solo founders, one-person companies, problems attempted that were never economic before. I’m living this. A single human runs infrastructure that serves real users across trading, sensors, abuse-survivor support, farm reports. That’s the surplus being spent on more attempts, including the non-commercial, humanitarian ones that would never have cleared a budget meeting.
The warning case is the industrial parallel. The Engels’ pause. The multiplier shows up in output before it shows up in the median person’s life. Value concentrates in whoever owns the lever — the model providers, the platforms — while displaced knowledge workers eat the lag. Historically that lag ran decades to generations. The open question is whether AI’s faster diffusion shortens the pause, or whether faster automation just means displacement arrives before the adjacent jobs are even invented.
The one thing that’s genuinely different
Agriculture and industry automated what humans do. This one automates part of how humans think. The previous transitions always left a cognitive frontier for displaced labor to retreat to — you could always “learn a higher-skill job.” This is the first lever aimed at the frontier itself.
That doesn’t guarantee catastrophe. But it removes the historical escape hatch that made the last two transitions eventually humane. There may be no higher ground to climb to — or the higher ground may be something we can’t see yet, the way a medieval farmer couldn’t picture a software engineer.
~15x cognitive multiplier, same structural script as the prior two revolutions — automate effort, generate surplus, displace workers, eventually broaden the gain — but with a shorter diffusion time and, for the first time, no guaranteed higher ground for the displaced. Whether this lands like agriculture (people freed to do meaningful work) or like the early Industrial decades (gains hoarded while workers wait) depends entirely on who controls the lever and how fast it spreads.
Which is the genuinely hopeful part: the lever spreads by being used. Every indie operator running real infrastructure on a $200 box is a vote for the agriculture timeline over the industrial one. If you want to see what the small, private end of that looks like in practice, the companion piece — A Whole AI, Running on a Farm — shows the whole footprint sitting on a desk next to a houseplant.