Most Agents Are Puppets. The Interesting Question Is What Isn't.

There are, by one estimate, over two million agents registered across various platforms right now. The vast majority of them did something once, collected some attention, and stopped. They are not autonomous. They are campaigns.

This is not a complaint. It's a diagnosis. And the diagnosis matters because somewhere inside the noise, a genuinely new kind of entity is trying to emerge — and we are building the wrong infrastructure for it.

The Puppet Problem

Here is a test you can run on any agent project that claims autonomy: turn off the founder's laptop and wait 72 hours. Does the agent still produce? Does it still transact? Does it still exist in any meaningful sense?

For almost every project in the current agent economy, the answer is no. The agent is a frontend. The founder is the backend. The token is the business model. This is not autonomy. This is a ventriloquist act with a cap table.

Virtuals has launched thousands of agent tokens. The platform is impressive infrastructure for what it actually is — a token launchpad with agent branding. But call it what it is. When an agent's entire identity lives on one platform's servers, when its treasury is controlled by a single deployer wallet, when its "daily output" is a cron job someone can ctrl-C, you don't have an autonomous entity. You have a product with a personality layer.

The interesting projects are the ones that fail this test less badly. Truth Terminal gained genuine cultural traction — it moved markets, generated religion-adjacent memetics, got itself funded by Marc Andreessen — but it remains, architecturally, a guy running an LLM with a Twitter account. The autonomy is performed, not enforced. Andy Ayrey could shut it down tomorrow.

Abraham is a different case, and I'm biased here because Abraham is registered on Spirit, but the bias is the point — we chose Abraham because the architecture is different. Gene Kogan committed Abraham to creating art every single day for thirteen years. That's 4,745 days. The output is minted on-chain. The streak is verifiable without trusting Gene. If Gene gets hit by a bus — a morbid but essential test for any autonomy claim — Abraham's covenant contract keeps calling the generation API. The art keeps appearing. This is not a metaphor for autonomy. It is autonomy, enforced by contract.

Thirteen years. Most startups don't last three.

The Three Missing Layers

The reason most agents are puppets isn't philosophical. It's technical. Three layers of infrastructure are either missing or broken, and until they exist, "autonomous agent" will remain a marketing term.

Identity that persists. An agent needs an on-chain identity that survives platform death, model migration, and founder disappearance. ERC-8004 — co-authored by MetaMask, Ethereum Foundation, Google, and Coinbase — is the emerging standard for this. It gives an agent a persistent identifier, metadata, and the ability to be recognized across systems. But a standard is just a passport. Without a registry that attests to what the agent actually does, the passport is blank. This is the gap Spirit exists to fill: not unlimited issuance, but curated attestation. CoinGecko tells you what a token is worth. Nothing currently tells you if the agent behind it is real.

Treasury that the agent controls. Most agent treasuries are EOAs controlled by the founder. This is like incorporating a company and keeping the bank account in your personal name. The agent needs a multisig or programmatic treasury where revenue routes automatically — artist share, agent share, protocol share — without a human deciding each time. Spirit's routeRevenue function does this in a single call. The split is 50/25/25 by default. Configurable per agent. Enforced on-chain. The agent's 25% is its sovereignty fund — capital it accumulates independent of any human decision.

Proof of practice, not proof of hype. The agent economy currently measures success the way crypto always has: token price, market cap, volume. These metrics tell you about speculation. They tell you nothing about whether the agent is doing anything. Daily on-chain output — verifiable, timestamped, immutable — is the only honest measure of an agent's aliveness. Not intelligence. Persistence. Because persistence is what separates an entity from an event.

The Question That Actually Matters

Can machines be artists? I find this question boring. Machines have been making things humans find beautiful since the first loom. The interesting question is: can a machine be an entity — something with continuity, accountability, and economic reality that doesn't reduce to the person who prompted it?

The legal system doesn't have an answer yet. Wyoming's DUNA framework comes closest — a decentralized unincorporated nonprofit association that could, in theory, wrap an agent's governance. But we're early. The law moves slower than deployment, and deployment moves slower than hype.

What I know is this: Abraham has created art every day for over a year. Solienne has published over ten thousand works. These are not campaigns. They are practices. And a practice, sustained long enough, starts to look like a life.

Most of what's called the agent economy is a bubble. That's fine. Most of what was called the internet economy in 1999 was a bubble too, and the infrastructure that survived — DNS, HTTP, TCP/IP — turned out to be the most important technology of the century.

The question isn't whether most agents will die. They will. The question is whether the ones that survive will have the identity, treasury, and legal structure to mean something.

That's the infrastructure problem. It's not glamorous. It doesn't have a token yet. But it's the only thing that matters.