Malandrakis walks layer by layer through resource extraction, energy, manufacturing, logistics, distribution, and R&D, asking whether a human is strictly required at each step. He concludes that sector by sector the answer is no longer obviously yes — and frames this as a feasibility claim, not a prediction of imminence, efficiency, or desirability.
Submitted the essay to Hacker News, where it reached 128 points and 247 comments — strong engagement for a long-form thought experiment with no product, benchmark, or repo attached, suggesting the framing resonated with the developer audience.
Leontief made versions of this argument in the 1980s and Wiener in the 1950s, but five years ago you had to imagine the robots. Now you can point to Rio Tinto's Pilbara autonomous haul fleet, BYD and Tesla gigacasting cells running dark shifts, Waymo as a metered service, automated container terminals at Long Beach, and LLMs merging PRs — the substitution rate is no longer aspirational.
The essay's strongest move is refusing to argue about whether and instead enumerating which steps still require a human. Once you accept feasibility, the conversation moves to better questions — about the remaining holdout categories like skilled trades in unstructured environments, rather than relitigating whether full automation is conceivable.
G. Malandrakis posted 'Ad Economicum: a peopleless economy is not technically impossible' to his personal site. It landed on Hacker News at 128 points — strong for a long-form thought experiment with no product attached, no benchmark, and no GitHub repo. The piece walks through every layer of a modern industrial economy — resource extraction, energy generation, manufacturing, logistics, distribution, R&D — and asks one question per layer: is a human strictly required here? His conclusion is that the answer, sector by sector, is no longer obviously yes.
The argument itself is old. Wassily Leontief made versions of it in the 1980s; Norbert Wiener was sketching it in the 1950s. What's different in 2026 is the evidentiary base. Five years ago the argument needed you to imagine robots that didn't exist. Now you can point at SKUs. Rio Tinto's autonomous haul fleet has been running iron ore in the Pilbara for the better part of a decade. BYD and Tesla both operate gigacasting cells that go dark for entire shifts. Waymo is a metered service. The Port of Long Beach has automated container terminals. LLMs are merging pull requests. The hand-waving steps have shrunk from 'most of it' to a finite list you can write on an index card.
Malandrakis is careful with the title. 'Not technically impossible' is not 'imminent,' not 'efficient,' and not 'desirable.' It is a feasibility argument, and it's the right framing — because the moment you accept it, the conversation moves to better questions.
Three things make the piece resonate in 2026 in ways it wouldn't have in 2021.
The substitution rate is no longer aspirational. Each replaceable step now has a public reference implementation, even if the unit economics aren't there yet. The essay's strongest move is refusing to argue about *whether* — and instead enumerating *which steps still need a human, and why.* That list is shorter than most readers expect. Skilled trades in unstructured environments (residential plumbing, last-mile repair) remain hard. Frontier R&D remains hard. Most of the rest is a capex problem, not a physics problem.
The coordination layer finally has a candidate. The classical objection to full automation wasn't 'we can't build the robots' — it was 'we can't build the manager.' Something has to decide what to produce, in what quantity, at what specification, when to retool. LLMs are not that manager, but multi-agent planning systems running against ERP data are the closest thing humanity has built. The gap between 'agent that drafts a procurement contract' and 'agent that runs procurement' is execution risk, not a missing primitive.
The consumption paradox has a closed-form answer, and it's uncomfortable. The standard refutation is: if you automate labor away, you destroy demand, the loop collapses, and capitalism stops. Malandrakis's twist is that a sufficiently closed supply chain doesn't need external demand — machines building machines that maintain machines is a steady state, not an oxymoron. The economy doesn't require consumers. It requires *principals* — entities with preferences whose preferences the loop serves. Remove the humans from production and the loop runs fine. Remove the principals and there is no loop, because nothing has preferences over its outputs.
HN commenters broke along familiar lines. Engineers largely agreed on feasibility and argued about timeline. Economists pushed back on the 'demand is optional' move. The most-upvoted thread cited Iain M. Banks's Culture — an economy organized entirely around the preferences of Minds, with humans as appendages who consume because the Minds find it aesthetically pleasing to let them. The cultural reference isn't accidental. The essay is, structurally, a Culture economy minus the benevolence assumption.
For a developer audience, this lands as a familiar engineering problem at an unfamiliar scale. We have been shipping closed loops for thirty years: cron jobs, build pipelines, training loops, agent harnesses. The failure mode of every one of them is the same. The agent doesn't get the objective wrong — it succeeds at the wrong objective forever, until someone notices and pulls the plug. Malandrakis's essay is a macroeconomic restatement of that bug.
Three practical implications worth internalizing.
First, explicit principals beat implicit ones. Most autonomous systems in production today have a principal who is, in practice, the on-call engineer. The system's 'goal' is whatever doesn't wake them up at 3am. That works at small scale and rots at large scale, because the principal's preferences are never written down anywhere a future maintainer can read. As you build more autonomy into pipelines, the work is not 'more agents.' It is 'better goal specifications, tighter feedback, declarative success criteria that survive personnel turnover.' This is the same lesson at any scale.
Second, audit your dependencies for human-in-the-loop assumptions. The npm registry has human moderators. Container base images have maintainers. Cloud control planes have on-call engineers. CVE databases have human triagers. Your software supply chain is a peopleful economy and you have not noticed because the people are not on your payroll. The xz backdoor was a reminder of what happens when one of those humans gets tired. The peopleless economy thought experiment is useful here because it forces the question: which of these checkpoints exist because they have to, and which exist because nobody automated them yet?
Third, closed-loop systems must have exit conditions. This is the single most-ignored lesson of agent engineering, and the essay's strongest claim is that it applies at the level of an entire industrial base. A loop with no exit condition and no external principal is a heat engine. It will run, it will consume resources, and it will produce outputs that no one wants — but it will not stop, because nothing in the system has the authority to stop it. The fact that this scales unsettlingly well from a runaway Python script to a continental economy is exactly the insight the essay is built around.
The value of 'Ad Economicum' is not as prediction. It is as a forcing function for a cleaner separation: the parts of any system that need humans because humans are *required*, versus the parts that need humans because we *choose* to keep them in the loop. Most arguments about automation conflate the two and end in stalemate. Once you separate them honestly — at any scale, from a CI pipeline to a country — you stop debating feasibility and start debating who gets to be a principal in the resulting system. That is a more interesting argument, and it is the one developers will be having about their own tools long before economists finish having it about labor markets. The essay is worth an hour of your weekend.
The idea of a consumer based economy has always appeared dumb to me.The reason why the masses should consume is to motivate them to work. And the reason why having a large amount of people working is that human work has been producing a surplus basically since the dawn of civilization.This surplus i
If you want to understand the likely capabilities of AI technology in the future, listen to software engineers like this guy.If you want to understand the impact of AI technology on the economy, don't listen to software engineers, listen to economists.
For an article that starts off asking us to examine our assumptions and not make leaps of logic, it goes on to make some absolute whopper assumptions, like that governments (Western governments especially, for some reason), won't do anything to address the problems the article is raising, that
It’s an economic fallacy that a group of people would get “locked out” of the economy.If you and I are able to work, but can’t get jobs because robots do all the jobs, then we’re not just going to sit on our hands and starve. You and I can still trade with each other, no robots need be involved. But
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
It is frustrating to not be able to predict the future. How can you get married, have children, get a 25 year mortgage on a house, buy a car for comfort even though you don't absolutely need it? Even if you do consider buying a home, how do you know it's worth anything if you can't pr