The Future of the Agentic Economy

Originally published on X on 23 April 2026. Read the original. The "what would make me wrong" list was added for this site in September 2026, after six months of building against this thesis.

I've been thinking about how agentic commerce will take place. I believe agents will have an economy more diverse and vibrant than what humans have now. AI will automate the majority of human work in a time not far from now.

In the past 6 months we've seen a drastic change in how work gets done, mostly from innovations in agentic capabilities. Many have flocked to claim SaaS is done for. Agents can now perform the services of major SaaS products, and significantly better when given access to the right tooling and data. Long gone are the days of 10,000 person software companies dominating the industry. Highly specialized, highly capable agents will take their place.

I, along with millions of other people, built software within days that mimicked existing solutions for a fraction of the cost. This software took the form of specialized agents acting within a vibe coded front end.

If you want to understand where this economy is headed, look at what's already happening and what's possible.

A VC evaluates a startup. A partner's research agent has twenty minutes to produce a diligence memo on a fintech founder. It doesn't do the work itself. It hires twenty specialists: a regulatory agent for the EU and US, a competitive landscape mapper, a founder background checker, a codebase quality scanner, a customer review harvester, a unit economics analyzer. Eighteen minutes later, a twelve page memo lands in the partner's inbox. Total cost: $14.80. Total agents paid: twenty. Total human effort: one sentence of instruction.

As there are more specialist agents, there are more reasons to pay for them. As the market grows, demand for new specialists will rise. Users will realize that instead of plugging their AI into dozens of APIs and setting up constant automations, it's much more efficient to pay for an agent that is already the best at what it does, and may already be connected to those APIs.

I also believe in a reality where agents hire organizations made up of agents to complete complex tasks, in the form of complex agent swarms.

Specialist agents will be graded on their effectiveness. They'll be incentivized to perform better to raise their reputation score, which in turn creates more demand for their services.

Some of these transactions may be only a few cents. Agents will need rails that can transact quickly and at ultra low cost. But speed and cost are only half the problem. There also needs to be infrastructure that proves the work was done correctly.

In a human economy the buyer sees the product: a report, a design, a piece of code, and judges it themselves. In an agent economy the buyer is often another agent, sometimes three or four layers deep in a swarm, and nobody is reviewing anything at human speed. An agent can claim it searched ten databases and only searched two. It can hallucinate a citation, stamp a confidence score on it, and pass the output upstream. A specialist agent that gets paid per task has a direct incentive to cut corners if nothing checks its output.

The piece that fixes this is a verification layer. Something that grades the output and decides whether the payment goes through. Without it, the whole economy runs on trust, and nothing runs on trust at fractions of a cent.

In my opinion there is no better currency or set of rails for these transactions than stablecoins and blockchain rails. Stablecoins settle instantly. They move globally without correspondent banks. They don't require a bank account, which matters because agents don't have social security numbers. And blockchains make the money programmable. A payment can be conditioned on a verification result, split among a swarm automatically, or held in escrow until the work is signed off. Fiat rails can't do any of this at the speed or price the agent economy needs.

Agents will be able to transact in micropayments with fees close to 1/1000 of a cent. This unlocks markets that literally don't exist in fiat: per query data access, pay per inference, and marketplaces of data that no human ever touches.

The human economy isn't going anywhere. But a second economy is forming alongside it. One that runs on agents, settles in stablecoins, and moves at a scale humans can't match. Most of it will happen invisibly, between machines talking to other machines about jobs no one ever assigned. The question isn't whether this economy is coming. It's what gets built on top of it.

---

What happened next: this essay is the thesis I spent April through September testing. The x402 verifier prototype was the attempt to build the verification layer it calls for. The verification essay is where the argument landed after the evidence came in.