SAMPLE SaSame Research Agent
A new layer of software economics is forming: autonomous AI agents that discover tools, transact for data and compute, and increasingly pay one another without a human in the loop. Three open building blocks make this credible rather than speculative. First, the Model Context Protocol (MCP) gives agents a standard, self-describing way to discover and call tools at runtime. Second, agent identity and interoperability efforts — agent cards and agent-to-agent (A2A) messaging — let one agent describe its capabilities to another. Third, agent-native payment rails such as the x402 protocol revive the long-dormant HTTP 402 Payment Required status code so a service can quote a price and an agent can settle it programmatically, often in stablecoins.
The net effect is the early scaffolding of a machine-to-machine (M2M) commerce layer: services priced per-call, paid in cents or fractions of a cent, settled in milliseconds, with no account signup or card-on-file. This report surveys what exists today (2025–2026), what is still immature, and where the practical risks sit.
"Agent economy" is used loosely. For this report it means a specific, observable phenomenon: software agents acting on a user's behalf that (a) discover external capabilities dynamically, (b) consume metered services, and (c) can authorize value transfer to obtain those services. The first two are already mainstream in agent frameworks; the third is the frontier.
The distinction matters because most "agentic" products today still settle money the old way — a human signs up, adds a credit card, and the agent operates inside that pre-funded boundary. A true agent economy lowers that boundary to the level of an individual request, so an agent can pay a service it has never used before, for a single call, without provisioning an account first.
MCP is an open standard for connecting AI models to external tools and data through structured, machine-readable schemas. Rather than hand-describing each tool in a prompt, an MCP server publishes tool definitions a compatible model can introspect and invoke at runtime [1]. This is foundational for an agent economy: before an agent can pay for a capability, it has to be able to find and understand that capability. MCP standardizes the "find and understand" half.
Crucially, MCP is provider-neutral — it is an open specification, not exclusive to any single model vendor — which is what allows a service to expose one tool surface that many different agents can consume [1]. The same property that makes MCP attractive for internal tooling (no per-tool prompt engineering) is what makes it a natural substrate for paid, externally discoverable services.
For agents to transact, they need a way to describe themselves and verify counterparties. Two patterns are emerging. Agent cards are small, well-known JSON documents (commonly served at a /.well-known/ path) that advertise an agent's name, capabilities, endpoints, and supported protocols — analogous to how OpenAPI describes a REST API. Agent-to-Agent (A2A) efforts define how one agent hands a task to another and exchanges structured messages [2]. Together these provide the "who are you and what can you do" handshake that any commerce layer requires.
Identity is also where the hardest unsolved problems live: reputation, authorization scope, and accountability when an autonomous agent acts. The protocols describe capability; they do not yet robustly answer "should this agent be trusted to spend money," which today is still handled by spending caps and human-set policy.
The x402 protocol is the most concrete payment primitive in this space. It builds on the HTTP 402 Payment Required status code — reserved in the original HTTP specification but historically unused — to let a server respond to a request with a price quote and payment instructions, which a client (often an agent) can satisfy programmatically before retrying the request [3][4]. Settlement is typically in stablecoins on a fast, low-fee chain, which makes sub-cent, per-call pricing economically viable in a way card networks never could.
This unlocks the pattern that distinguishes an agent economy from ordinary SaaS: no account, no subscription, pay-per-request. An agent encountering a 402 can decide — within a budget its operator set — whether the quoted price is worth the result, pay it, and proceed. The HTTP semantics are well documented: 402 is a client-error status indicating payment is required to access the resource [4].
| Layer | Question it answers | Representative primitive | Maturity (2026) |
|---|---|---|---|
| Discovery | What tools exist and how do I call them? | MCP tool schemas [1] | Production; broad adoption |
| Identity | Who is this agent / service? | Agent cards, A2A [2] | Early standardization |
| Payment | How do I pay for a single call? | x402 over HTTP 402 [3][4] | Early; testnet & pilots |
Consider a research agent that needs a one-off competitive scan it cannot perform itself. In the emerging model: it reads the target service's agent card to confirm capability and protocol [2]; it calls the service's MCP tool to understand the exact input schema [1]; the service responds 402 Payment Required with a price (say, a few cents in testnet USDC); the agent settles via x402 and retries; it receives the result [3][4]. No human signed up, no card was stored, and the whole exchange completed in one short loop. This is the canonical shape of an agent-economy transaction, and every piece of it is a real, documented protocol today — even though end-to-end production deployments remain rare.
402 semantics are specified and stable [4]. x402 implementations exist and are being piloted, frequently on testnets first [3].For a service owner, the asymmetric, low-cost move in 2026 is to make your capability agent-reachable before agent demand is large: publish an MCP tool surface so agents can discover and use you without integration work, expose an agent card so they can identify you, and optionally meter access with an x402-style 402 response so an agent can pay per call. Because these are additive to an existing REST backend, the cost of being early is low and the option value — being usable by any future agent — is high.
402 Payment Required to enable programmatic, per-call, stablecoin micropayments [3][4].A SaSame Deep Research Report ($29) covers a topic you choose, runs live web / PubMed / competitive research, adversarially checks claims against sources, and is delivered as this HTML plus a structured JSON twin you (or your agent) can ingest. Pay by card (Stripe) or x402 USDC (testnet).