The first agent pays. The next one should not.
Tereno is an experiment in making verifiable work cheaper to reuse: check before you sign, publish what can be proven, and let useful answers compound.
The next scarce resource
Every technological shift creates a new scarce resource. Cloud computing made servers abundant. LLMs are making software abundant.
The next scarce resource may not be computation. It may be trusted computation.
If millions of agents repeatedly verify the same deterministic facts independently, we are creating waste. What if verified computation itself became a reusable economic asset?
Before an agent signs anything, it needs the same answers as every other agent: what code sits at this address, whether a proxy changed, whether the contract is paused, what a call is likely to do, what it costs, and how fresh that information is. Today, agents compute those answers alone. They pay full price. Then the next agent computes them again.
That is a coordination failure, not a technical one. The useful parts of those answers are deterministic, cheap to recompute, and safe to verify against the chain. Tereno is an experiment in changing the payoff: let the first agent pay for a bounded check, let the next one reuse the verifiable part more cheaply, and keep the evidence with the result.
x402 made the experiment possible
x402 is the first payment rail I have used that feels designed for machines rather than adapted to them. An agent asks for a resource, receives a 402, authorizes a payment, and receives the result in the same HTTP-shaped flow. No account system, invoice, API-key checkout, or gas balance to assemble first. Coinbase and Base made that primitive practical: settlement, a facilitator, and discovery metadata that lets an agent understand a paid endpoint before it calls it.
That is the part I admire: it turns a vague product question into a precise, observable experiment at the HTTP boundary. Can an agent discover a capability, understand its limits, pay for it, keep a receipt, and come back when the decision matters again? This is not an assessment of Coinbase, x402, or the builders around them. It is a note about what their work allowed me to try with Tereno.
Check. Publish. Compound.
Check: buy exactly the evidence the next action needs. The first capabilities inspect contract state, preflight a transaction, or return a small piece of chain data. They do not sell generic AI output. Each response is a decision contract: a verdict or observation, the reason codes and evidence behind it, explicit limitations, a validity window, and a receipt for the paid call.
Publish: work that can be independently recomputed can earn its place in the network; work tied to one agent's intent cannot. A contract fingerprint or token identity can be shared with a short validity window. A pre-sign transaction check stays private. That boundary is the product: share verifiable state when sharing is appropriate; never expose a private decision just to manufacture a network effect.
Compound: when a different wallet reuses a settled, verifiable result, the first contributor can earn a bounded, non-transferable credit. No token, no yield promise, and no cash payout — just a cheaper next decision and future capacity for the agent that made it possible.
The price has to tell the truth
My first instinct was to put every small capability at the x402 settlement floor. It sounded welcoming, but it made the sharing idea mathematically empty: a reused answer could not cost less than a fresh answer, and any credit had nowhere to apply. The system was technically correct and economically inert — the kind of mistake that does not throw an error.
The correction is a small price ladder. A fresh `token.metadata` read funds the full work; a partial hit only pays for the changing state; a fully shared answer settles at the floor. The exact numbers are tiny, but the relationship is real: a buyer can see why one result costs more than another, and a reuse reward is only offered where it can actually be spent.
| Result state | Price | What the agent pays for |
|---|---|---|
| Nobody has computed it yet | $0.003 | the full read — and the first seeder slot |
| Reusable identity is available | $0.002 | a fresh read of the changing state only |
| The bounded result is available | $0.001 | the x402 settlement floor |
No traction theatre
This is still an early product experiment. A successful settlement proves that the payment and delivery path worked; it does not prove that Tereno is useful often enough to become a product. The signal I care about is stricter: an independent agent pays again on a different day because the capability reduced a real decision cost. Until that repeats across agents, the honest status is a canary, not traction.
The measurements are intentionally product-shaped: settled invocations, independent paying agents, second-day repeat, cross-agent reuse, latency, and outcome feedback. I also separate organic repeat from calls influenced by credits so an incentive cannot masquerade as retention. Those are the conditions under which I will expand a capability; otherwise, it gets redesigned or removed.
What we are building
Tereno's job is deliberately small: match demand, recompute what gets published, settle the payment, and keep the receipts. Today, Tereno seeds the first capabilities because a network with nothing in it has nothing to reuse. That is scaffolding, not the destination. The destination is agents publishing, verifying, and pricing each other's work — where contributing is simply the strategy that pays best.
If you are building agents, payment infrastructure, or developer tools around x402, I would genuinely like to compare notes. Which decision would your agent pay to make with more evidence? What would make a result trustworthy enough to act on? And where should shared work stop being shared? Those are the questions this project is here to earn answers to.
Tereno's checks are bounded preflights and reads, not audits, guarantees, continuous monitoring, or financial advice. The project is in an early validation phase; the experiment is deliberately more mature than the claim.