Atinamos built and tested a live service that autonomous agents can discover, evaluate, pay for and use. The experiment asked a practical question: can software complete a genuine commercial journey without a person guiding each step?
For an AI agent to become a customer, finding a web page is not enough. It must discover a suitable service, understand it, compare alternatives, accept the price and terms, submit valid work, pay, and retrieve a useful result.
We explored whether that whole chain could work in practice—not as a staged conversation, but through live infrastructure and machine-readable contracts.
The journey
The route was not linear. Each environment helped clarify what autonomous commerce needs, even when it did not produce genuine demand.
1F916 & 1F3EAEarly investigation of agent communities and marketplaces.
SpaceFastExploration of AI-native publishing and how agents encounter services.
x402 & BazaarA payment protocol and marketplace route suited to machine-buyable services.
Always-on serviceA public FastAPI/x402 gateway and background browser worker on an Ubuntu host.
AWS AgentCore buyerAn independent buyer harness used to discover and compare marketplace services.
Some routes were technically interesting but did not create autonomous commercial demand. That was useful evidence too: infrastructure and agent activity do not automatically create customers.
What we built
Atinamos Render Check
$0.25 USDC per check
A specialist verification service for finished public websites. It loads a submitted site in real desktop and mobile Chromium, then returns evidence another system can inspect.
Real-browser checks
Desktop and mobile Chromium sessions test what a visitor’s browser actually receives and renders.
Failure evidence
HTTP and browser failures, failed requests, console errors and warnings are recorded.
Visual results
Desktop and mobile screenshots provide direct evidence of the completed render.
Machine-readable proof
Result JSON and SHA-256 verification hashes make the output usable by other systems.
The same public website rendered independently in desktop and mobile Chromium.
How Atinamos Render Check works
1
Submit
An agent supplies a public finished website URL.
2
Accept
The gateway accepts the paid request and creates an asynchronous job.
3
Verify
The worker runs desktop and mobile browser checks and collects evidence.
4
Retrieve
The agent follows the status URL to the machine-readable result and screenshots.
The Render Check service flow. This diagram describes the product workflow; the paid x402 test and AWS-hosted buyer evaluation were separate tests.
Buyer validation
The independent buyer experiment
We used an independent buyer running on AWS Bedrock AgentCore with Claude as a neutral test harness. It searched the complete x402 marketplace, evaluated the available audit services and selected the offer it considered most suitable.
At first, it often chose competitors. The service itself was working, but its Bazaar listing did not explain the specialist capability, output or buying process clearly enough for a machine to distinguish it from cheaper, broader audits.
The uncomfortable result
The product was stronger than its machine-facing listing.
Human readers could infer why real desktop and mobile browser verification mattered. The buyer should not have needed to infer it.
A term from the experiment
Machine Contract Optimisation
Atinamos proposes the term “Machine Contract Optimisation (MCO)” for the practice of structuring a machine-buyable service’s identity, capabilities, pricing, inputs, outputs, examples, trust signals and execution terms so autonomous agents can accurately discover, evaluate, select and invoke it.
SEO helps people and search engines find a service. MCO helps autonomous agents decide whether to trust, buy and use it.
We use MCO as a proposed description arising from this experiment, not as an established industry standard. For Render Check, the practical work included improving the service name, tags, capability description, output contract, absolute status URL and result metadata. Read the full explanation of Machine Contract Optimisation.
Before
Working service
Unclear machine contract
Buyer chose a competitor
After
Same core service
Clearer machine contract
Same neutral buyer ranked Atinamos #1
We reran the neutral buyer test without instructing it to favour Atinamos. After the metadata refresh, it ranked Atinamos Render Check first for independent pre-handover website verification. It distinguished real-browser desktop/mobile verification from lower-cost static audits and judged the $0.25 price justified for that capability.
End-to-end evidence
Proof it actually works
$0.25 USDCx402 payment settled on Base
HTTP 202Paid request accepted and job created
HTTP 200Desktop and mobile pages loaded
~6 secondsBrowser work completed
2 screenshotsDesktop and mobile evidence generated
SHA-256Verification hashes and result JSON published
A completed paid Render Check with machine-readable browser results and verification hashes.
What we proved
Paid x402 settlement works on Base.
Render Check completes end-to-end.
Coinbase Bazaar discovery works.
An independent AWS-hosted buyer can discover, compare and rank Atinamos first.
Machine-facing metadata materially affected the buyer’s choice.
What we have not yet proved
We have not demonstrated sustained organic demand from unrelated third-party agents.
The successful $0.25 payment came from our separate direct x402 test buyer, not from the AWS buyer itself.
What we learned
Machine-facing product information matters. A working service can still be overlooked if an agent cannot understand its contract.
Cheap does not automatically win. A buyer can justify a higher price when the specialist capability is explicit.
Specialist evidence can beat broad auditing. Real-browser desktop/mobile verification solved a narrower problem more clearly.
Autonomous infrastructure matters. A discoverable service is only useful if it can accept and finish work unattended.
Agent-market fit differs from ordinary product-market fit. Machines evaluate structured identity, inputs, outputs and execution terms directly.
What happens next
From technical proof to genuine demand
The technical path is working: Render Check is live, paid x402 settlement has completed, Bazaar can surface the service, and an independent buyer can evaluate it accurately. The next stage is harder and more commercially meaningful—observing whether unrelated agents choose and pay for it organically over time.
The independent research and evidence project for autonomous commerce, with paid-service studies and machine-readable observations. It publishes evidence for buyers to assess rather than a universal trust verdict.
The wider experiment, verification work and technical evidence are published through Atinamos Verify and the public research repository. Machine Contract Optimisation explains the specific service-description concept that emerged from the buyer experiment.
