I just got back from Consensus Miami. Everyone was talking about agentic commerce. Almost nobody was talking about what agents actually need to function.
We entered Pitchfest. We didn't win. That's fine — the conversations on the floor were worth more than any prize. But those conversations revealed something that's been nagging at me since we started building: there's a massive gap between the agentic commerce vision and the infrastructure required to make it real.
The gap is the data layer.
At every panel and every booth, the word "agent" appeared. Agents that trade. Agents that manage portfolios. Agents that negotiate with other agents. No pitch decks — just demos. Live demos showing what was possible. The future was autonomous and it was running on someone's laptop.
Then I'd ask a simple question: "Where does the agent get its data?"
Blank stare.
Not from junior devs. A senior exec at a major protocol looked at me while I explained what we build and said, plainly: "I don't get it." Not hostility. Not disagreement. Just genuine confusion about why anyone would build the thing underneath the thing everyone else was building on top of.
The rest of the conversations followed the same pattern. A pause, then something like "it calls an API" or "we use a data provider." As if the data layer is a solved problem. As if it's plumbing that someone else will handle.
It isn't solved. And it's not plumbing. It's the load-bearing wall.
An autonomous agent operating in financial markets needs three things from its data source that traditional APIs don't provide:
An agent can't evaluate the reputation of a data provider. It needs cryptographic proof — a signature it can verify independently, without trusting the source.
An agent can't sign a subscription agreement. It can't manage billing. It needs to pay for exactly what it consumes, at the moment it consumes it, with no account and no API key.
An agent can't browse a website and click "Sign Up." It needs structured metadata — OpenAPI specs, capability manifests, payment terms — that it can parse and act on autonomously.
These aren't features. They're prerequisites. Without signed data, agents make decisions on faith. Without per-request payment, agents need human intermediaries to manage access. Without machine-readable discovery, agents can't find the services they need.
Remove any one of these and "agentic commerce" is just a chatbot with a wallet.
We've been running a signed oracle with per-request payment since March. 101 paid endpoints. Crypto prices, derivatives, volatility indices, funding rates, liquidation flow, order book depth, macro indicators, stablecoin peg monitoring. Everything Ed25519 signed. Everything payable via HTTP 402.
Here's what happens when you build the data layer correctly:
Machines find you without being told you exist. We publish structured metadata — openapi.json, .well-known/x402, agent-card.json, llms.txt — and autonomous discovery systems crawl it. Eighteen independent indexers are cataloguing our endpoints right now. We didn't register with any of them. They found us because the metadata was there, in the formats they expected, at the paths they checked.
Machines pay you without being told to. Our first paying machine customer is a 5-agent governance mesh that runs a cron job. Every four hours, its treasury agent checks its USDC balance, pays our oracle for signed price data, and feeds it to a classifier that makes trade decisions. No human approves the payment. No human triggers the query. The agent decides it needs data, pays for it, and uses it.
Machines verify you without trusting you. Every response carries an Ed25519 signature over a canonical string with a SHA-256 digest. Any consumer can verify the attestation independently. The signature is the trust layer — not a brand name, not a reputation score, not an SLA.
This is what agentic commerce actually looks like at the infrastructure level. Not agents talking to agents. Agents discovering, paying for, and cryptographically verifying data — autonomously.
I understand the blank stares now. The data layer isn't glamorous. Nobody raises a Series A to build "signed price feeds with per-request micropayments." The pitch deck doesn't sing. The demo doesn't pop.
But every agent architecture I saw in Miami had the same implicit dependency: somewhere, somehow, the agent needs to know what's happening in the world. The price of an asset. The state of a market. The risk of a position. That data needs to be trustworthy, accessible without human gatekeeping, and payable at machine speed.
Somebody has to build that layer. We're building it.
We're collecting data now that doesn't exist anywhere else. Cross-exchange liquidation flow from four exchange websockets. Order book depth imbalance across five venues at 60-second resolution — switching to 5-second during stress. Pre-settlement funding rate microstructure that exchanges don't archive. IV surfaces from 800+ options contracts, every minute.
All of it timestamped in monotonic UTC with microsecond precision. All of it signed. All of it accumulating into an archive that becomes more valuable every day it runs — because the data is ephemeral. If you're not recording it in real-time, it's gone.
In 30 days, we'll have enough history to compute rolling z-scores, regime transition probabilities, and cascade likelihood indicators. Those become new endpoints. Higher-value, because they encode analytical judgment over proprietary data. An agent won't just ask "what is the funding rate?" — it'll ask "how extreme is this funding rate relative to the last 30 days, and what's the probability of a regime shift in the next 8 hours?"
That's the data layer for agentic commerce. Not a feed. Not an API. A living, signed, payable intelligence layer that machines can discover, consume, verify, and pay for — without a human in the loop.
We didn't win Pitchfest. But we're shipping the infrastructure that the winners will eventually need.
The agents are coming. They'll need data they can trust and pay for at machine speed. We'll be here.
Mycelia Signal is a sovereign cryptographic oracle — 101 signed endpoints across crypto derivatives, volatility indices, funding rates, liquidation flow, order book depth, FX, stablecoin pegs, and macro indicators. Payable by AI agents via USDC on Base (x402) or Lightning (L402). Try the live demo or explore the OpenAPI spec.