Digital Communism

There’s an old joke that’s been making the rounds again:

“The factory of the future will have only two employees, a man and a dog. The man will be there to feed the dog. The dog will be there to keep the man from touching the equipment.”

It’s often attributed to Warren Bennis, but he didn’t come up with it. The earliest version appeared in a 1978 British trade journal where engineers at the Post Office were already joking about their own obsolescence.

I’m sitting at my desk right now building OpenClaw.rocks. Our dog Yoshi is lying next to me. I’m running Claude Code, and my primary contribution to this codebase today has been clicking “confirm.” I still write prompts. I still make architectural decisions. But every month, the ratio of my input to the system’s output shifts further in the system’s favor.

I am, quite literally, the man. And Yoshi is the dog.

This post is about what that means. Not just for me, but for software, for physical goods, and for the economy that produces them.

What I mean by digital communism

I should define the term, because I don’t mean it the way Marx did.

Classical communism is an economic system where the means of production are collectively owned and goods are distributed based on need. Nobody tried to make it happen through markets. It was supposed to require revolution.

What’s happening now is different. Market forces, not ideology, are producing communist-like outcomes for digital goods. The means of production for software, content, and design are becoming so cheap that anyone can own them. A laptop and an API key give you capabilities that cost millions five years ago. The outputs are approaching free. Distribution is global and instant.

Nobody seized anything. Nobody revolted. The cost of intelligence just collapsed.

Sam Altman told a Federal Reserve conference in July 2025 that AI inference costs have been dropping by 10x per year for five years straight. In his essay The Gentle Singularity, he writes that “intelligence too cheap to meter is well within grasp.” That phrase echoes Lewis Strauss in 1954 promising that nuclear energy would make electricity “too cheap to meter.” That prediction was wrong. But unlike nuclear energy, compute costs actually are falling at the rate Altman describes. And unlike a reactor, you don’t need a government license to use an LLM.

When intelligence becomes effectively free, everything intelligence can produce follows it down. Software. Content. Design. Legal documents. Marketing. Code review. The raw material of the digital economy is becoming abundant in a way that looks, from the consumer’s perspective, a lot like communism. Free, available to all, distributed on demand.

I’m calling it digital communism. Not because it is communism, but because the outcome for consumers is converging on what communism promised: abundance without scarcity.

Why nobody will resist this

I’ve never cared about the assembly code my compiler produces. I write TypeScript, and trust that something turns it into instructions a processor can execute. I don’t inspect it. I don’t want to.

My mum has never cared about the code behind the software she uses. She opens an app, it does a thing, she’s happy. The implementation is invisible to her and always has been.

This is a pattern as old as computing itself. Each generation abstracts the one below. Assembly abstracted machine code. C abstracted assembly. Frameworks abstracted languages. Cloud abstracted servers. Nobody mourned the loss of direct hardware access. People chose the more convenient option every single time.

AI is the next layer. It’s already abstracting code for me. I describe what I want, an agent writes it. I think it will abstract applications next. Satya Nadella said on a podcast that the notion of business applications could collapse in the agent era. You won’t download a calorie tracker app. You’ll tell your agent to track calories. You won’t configure a project management tool. You’ll tell your agent to manage your project.

Clayton Christensen called this “Jobs to Be Done”: people don’t buy products, they hire them to do a job. They don’t care how the job gets done. They never have.

Digital communism isn’t just a supply-side story about costs dropping. It’s a demand-side story too. People will happily let go of every layer they never wanted to manage in the first place. This isn’t a tech trend. It’s human nature.

Three layers of scarcity

The reason I say “digital” is that this only works for things made of information. The economy has three layers of scarcity, and they’re being peeled away in sequence.

Intelligence is nearly solved. LLMs can write code, generate content, analyze data, and make decisions that used to require expensive human expertise. The cost is plummeting and the capability is rising. This layer is approaching free. It’s not truly zero, of course. AI still needs chips and electricity. But the cost of solar energy has dropped 89% in a decade, with the U.S. Department of Energy targeting $0.02/kWh by 2030. Even the material floor under intelligence is sinking.

Labor is next. In December 2025, Tesla posted a video of Optimus jogging in the lab at nearly 9 mph, with a true flight phase in its gait. A month later, Musk announced on Tesla’s Q4 2025 earnings call that Tesla is ending Model S and X production at Fremont and converting those factory lines to build Optimus robots instead, with a million-unit production line in the works. That’s a car company deciding robots are a better use of its factories than cars. Figure AI raised a billion dollars to build humanoid robots at their dedicated BotQ facility. Their Figure 02 just finished an 11-month deployment at BMW, loading over 90,000 parts with millimeter precision across 1,250+ hours of runtime. Hugging Face launched LeRobot, an open-source robotics framework with 21,000+ GitHub stars, and they’re selling open-source humanoid robots for $3,000. In China, UBTECH’s synchronized robot army went viral, XPeng’s IRON moves so realistically its CEO cut it open on stage to prove it wasn’t a person, and dozens of humanoid models are appearing simultaneously. I don’t think we’re more than ten years from robots that are economically viable for common physical work.

Materials are the wall. You can’t copy an atom. Even when intelligence is free and robots do the labor, you still need steel, lithium, copper, rare earths, and energy. Material scarcity is a fundamentally harder problem than the other two. It might require breakthroughs we can’t predict yet: asteroid mining, molecular assembly, fusion energy. That wall could hold for decades or centuries.

This is why the “digital” qualifier matters. For pure digital goods, we’re already approaching post-scarcity. For physical goods, free intelligence and free labor will drive costs down significantly, but materials set a floor that doesn’t exist in the digital world. A robot can assemble a car for free. The steel still costs something.

What the transition looks like

So we’re heading toward digital communism for information goods. What does that look like in practice, right now, for someone trying to build a software business?

It looks like the restaurant industry.

In early February 2026, Anthropic released new AI tools that triggered a sell-off wiping nearly $300 billion from software stocks. The software industry’s forward P/E ratio dropped to roughly 21x, down from 39x eight months earlier. SaaS companies that boasted 85% margins are adjusting to 60-70%. Analysts are calling it a SaaS apocalypse.

The structural problem: AI turns software costs from “per customer” to “per action.” When ten AI agents do the work of a hundred sales reps, you don’t need a hundred Salesforce licenses.

This is what restaurants have always dealt with. The ingredients are cheap. Anyone can open a kitchen. A few big chains like McDonald’s leverage economies of scale to maintain healthy margins. Everyone else operates on razor-thin ones. Many close within a year.

Software is entering that same phase. I can see it from where I’m sitting. In the OpenClaw hosting space alone, new competitors are appearing weekly. ClawSimple, ShootClaw, Quick Claw, PlugAndClaw, and others have all launched in the past few weeks. Browse TrustMRR or Product Hunt and you’ll find even more that haven’t been shared there yet, including Kilo Claw from a more established platform. AI made it possible for each of these founders to build and ship a hosting product in days, not months.

This is genuinely good for users. More competition, better prices, more choice. I support it even when my competitors are appearing faster than I can count them.

But here’s the honest question: if digital communism is the destination, and the gastro phase is just the transition, why am I building a business at all? If the outputs are heading toward free, what exactly am I selling?

The taste question

In restaurants, the ingredients are cheap but the meal is not. The value isn’t in the raw materials. It’s in the chef’s decisions. What to cook, how to combine it, what to leave out. Taste. Judgment. Curation.

The restaurant industry survives despite cheap ingredients because human labor can’t easily be automated. A robot can flip burgers at McDonald’s. It can’t run a kitchen that people travel across town for. Not yet.

In software, that distinction is collapsing. AI can already build the product, deploy it, monitor it, handle support. The “cooking” is being automated alongside the “ingredients.” So the gastro analogy has an expiration date for software. Restaurants are a stable equilibrium because the human element resists automation. Software gastro is an unstable phase. It’s heading further toward free.

Which means: for software businesses in this transition, the only durable value is the thing AI can’t do yet. Not building. Not operating. Deciding what’s worth building, and whether the result is good.

This connects directly to the slop question.

Merriam-Webster made “slop” their 2025 Word of the Year: “digital content of low quality that is produced usually in quantity by means of artificial intelligence.” It defined 2025. I think it will define 2026 even more.

Slop is what you get when AI produces without human taste. When nobody decides what’s good, when nobody filters, when the output is optimized for volume instead of quality. It’s the food poisoning of the digital restaurant economy.

Aaron Bastani calls the optimistic version Fully Automated Luxury Communism. Dario Amodei at Anthropic wrote about it in Machines of Loving Grace, predicting AI could compress a century of progress into a decade. The question isn’t whether we get abundance. We will. The question is whether abundance means quality or noise. Luxury communism or automated slop.

The answer, I think, depends entirely on whether a human with taste is still in the loop.

The distribution question

There’s a harder question underneath all of this that I’ve been avoiding.

If digital goods become free and physical labor gets automated, but materials still cost something, who has money? If AI collapses the value of most work, the IMF’s research suggests what we’d expect: wealth concentrates among those who own the AI and the capital, while everyone else’s skills lose market value. Those people might be able to afford everything digital, since it’s free. But steel, lithium, housing, energy? Those still cost something. And if you’re not earning, you can’t pay.

This is the gap between digital communism and actual communism. Actual communism has a distribution mechanism: from each according to ability, to each according to need. Digital communism, as I’ve described it, has no such mechanism. The abundance is real, but so is the question of who gets to participate in the parts of the economy that still require scarce resources.

People are thinking about this. Sam Altman has floated the idea of universal basic compute: instead of universal basic income, everyone gets a slice of AI capacity they can use, sell, or donate. It’s one possible answer. There are others. None are proven.

I don’t have an answer either. I hope this leads to a future where people have more meaningful time with the people they love. Where the baseline of what everyone can access rises high enough that the material floor matters less. Where the abundance of intelligence and labor translates into broadly shared prosperity, not just concentrated wealth with free entertainment for everyone else.

But I don’t know.

The man’s actual job

What I do know is what I’m doing right now. I’m not writing much code anymore. I’m not manually deploying servers. AI agents handle more of that every month. My job at OpenClaw.rocks has shifted from engineering to something harder to name.

I decide what to build and what not to build. I decide when the output is good enough and when it’s slop. I look at five AI-generated options and pick the one that actually serves the user. I maintain opinions about what matters. These are taste decisions. And they’re the only decisions that AI consistently defers to a human on, because “good” is not yet a thing AI can define for itself.

The man in the joke isn’t there to touch the equipment. The dog makes sure of that. The man is there because the factory needs someone who cares about what it produces. Someone who can look at the output and say: this is good, this is garbage, this is slop.

That’s my job now. Not building the factory. Not running the equipment. Caring about the output.

Maybe that’s what platforms like this become in the long run. Not economically significant, but something closer to art. The domain, the name, these articles, the opinions behind them. A thing someone built because they cared, and that other people choose over the alternatives not because it’s cheaper or faster, but because a specific human thought about it and that shows.

Yoshi is lying next to me as I write this. He doesn’t care about software margins or SaaS corrections or humanoid robots. He’s comfortable. The system works. I click confirm when the work is good and reject when it isn’t.

I don’t know if this is communism. I don’t know if the gastro phase lasts a year or a decade. I don’t know when AI will develop taste of its own and make even this job unnecessary.

But right now, someone still has to care. That’s the gig.

This post is itself a case in point. I didn’t write it word by word. I described what I wanted to say, an AI drafted it, and I provided a few rounds of feedback on the argument and structure. The same process I described above: the equipment does the work, the man provides the taste. If you’re curious what that looks like in practice, that’s the kind of app we’re building.

If you want to follow where this goes, come along.