Agents Can't Be Your Community: The Economics of Fake Growth

A note on this piece: the ideas, arguments, and conclusions here are mine. I used AI to help edit and structure the writing — tightening the prose, organizing the sections, checking that the economic reasoning held together. Think of it as an editor with unusual stamina, not a co-author. The thinking is human. The typing got some help.

There's a quiet assumption spreading through tech right now: that AI agents can absorb the work of community engagement, developer relations, and ecosystem building. Spin up enough agents, the thinking goes, and you have contributors, advocates, and builders without the headcount.

The economics don't support it. Neither does physics.

Every AI output has a money input behind it. Money functions as quasi-energy storage — the essential input in any economic supply chain. AI doesn't change that. It's a tool within the chain, and it still requires people to allocate electricity, hardware, and instruction. There is no output without a spender.

This is where the agent-as-community model breaks down. If you deploy agents to be your community members, your contributors, and your builders, you occupy both sides of the transaction. You are simultaneously the business and the customer. That arrangement is structurally unsustainable: it's a closed loop that must consume progressively more capital to keep producing output. In trading, this pattern has a name — wash trading. Activity that looks like growth but generates no external demand.

Thermodynamics explains why the loop can't close.

A community is a highly ordered system — structured relationships, accumulated trust, shared context. In physics, ordered systems don't persist on their own. The Second Law of Thermodynamics states that entropy, or disorder, increases in any closed system unless energy is continuously supplied from outside. A community is no exception. It requires a constant inflow of external energy to maintain its structure, and in economic terms, that energy is real spending from real people.

An agent-only community strategy severs that external input. It becomes a closed system, recycling its own output as its own input. Physics already tells us what happens next: without external energy, order decays. What looks like activity is just the system burning its reserves to resist entropy — and that resistance gets more expensive the longer it runs. The cost curve steepens while the value curve stays flat. Eventually the fuel runs out, and the structure collapses.

There's a second principle at work here too. Landauer's Principle, formulated by IBM physicist Rolf Landauer in 1961, holds that information is physical — every act of computation has a minimum energy cost. Every comment, post, or reply an agent generates burns real electricity. When that output has no external utility, when no real human finds it valuable enough to pay for, the energy spent produces nothing but waste heat. We are, quite literally, burning capital to generate noise.

Real growth requires real customers. Even in a product built entirely around agentic use cases, someone has to authorize the spend for those agents to run. That someone is a customer, and customers have choices.

Which brings us to the actual competitive problem. In a market where a competitor can replicate your idea within a week and offer it cheaper, retention is the only durable advantage. Not feature parity — that's temporary. Not price — that's a race to the bottom. Retention comes from trust, context, and relationships that don't transfer when someone clones your repo.

This is precisely the function that community and developer relations serve. Not as a cost center to be automated away, but as the mechanism that converts a product into an ecosystem people choose to stay in. It is the external energy input that keeps the system ordered.

The pressure here is not hypothetical. The AI industry's capital structure depends on long-term growth projections, and those projections require customers who remain. Companies carrying enormous valuations against future revenue need sticky, retained users far more than they need impressive demo metrics. Churn is the threat that automation cannot solve, because automation is what makes products easy to switch away from.

None of this argues against AI adoption. It argues against confusing automation with value creation. Agents can handle throughput, triage, and scale. They cannot originate demand, and they cannot hold a relationship. Those remain human functions — and in a commoditizing market, they're the ones that compound.