Machine customer
What Is a Machine Customer?
How agents discover, evaluate, buy, use, and dispute products on behalf of people and businesses.
A machine customer is an authorized software agent that evaluates or purchases a product or service on behalf of a human or organization.
What is a machine customer?
A machine customer is an authorized software agent that evaluates or purchases a product or service on behalf of a human or organization. The agent may discover options, compare policy and price, obtain a bounded credential or budget, transact, use the resource, preserve a receipt, and report the result to its principal.
The near-term opportunity is not an independent consumer robot with an unlimited wallet. It is a business agent buying bounded, machine-readable resources such as data, compute, software actions, reservations, or other services with explicit identity and spending limits.
How does an agent decide what to buy?
An agent needs a legible offer: a precise capability, price or meter, eligibility, input requirements, policy, permission scope, expected output, evidence, refund or reversal path, and a callable way to act. Ambiguous marketing copy forces the agent to guess or send the decision back to a person.
The agent buying journey resembles the human one—discover, evaluate, trust, transact, use, and recommend—but each stage needs structured evidence. Identity, authority, receipts, logs, and dispute handling matter earlier because the buyer is acting for someone else.
What should a homepage show an AI agent?
A homepage should state what the business can do, who the service is for, what the accepted outcome is, how pricing works, what authority is required, which policies govern the action, and where a machine can obtain structured details or call the capability.
This does not require a separate invisible site for agents. The human-readable page should contain clear textual answers and crawlable links. Machine-readable metadata and APIs should match the visible offer rather than advertise capabilities the customer cannot verify.
Why does identity come before payment?
A merchant needs to know which principal the agent represents, what authority was delegated, how much it may spend, what it may buy, and who receives the receipt or liability. A wallet without principal identity makes authorization and dispute handling fragile.
Credentials should be scoped to a purpose and budget. The transaction should preserve the principal, agent, policy version, requested capability, amount, output, and reversal path so both sides can reconstruct what happened.
How should a business price for machine customers?
Use meters tied to the resource or accepted action: a verified record, completed reservation, delivered file, unit of compute, successful API action, or other bounded result. Publish enough of the meter and policy for an agent to compare the offer before spending.
Protect the business against loops and abuse with quotas, idempotency, spending limits, rate limits, replay protection, and clear failure semantics. A machine buyer can repeat a mistake faster than a human, so the commercial contract must be executable as well as readable.
What happens when both customer and support are agents?
Both sides need a shared record of the request, authority, action, evidence, and policy version. Normal cases may resolve automatically, but disputes must name the principals and provide a route to human review.
The goal is not to remove people from accountability. It is to let agents handle routine, reversible work while preserving enough evidence for a person to understand and correct a contested outcome.
Field notes for this topic
Read the archive →The First Machine Customer Buys a Browser Session
Stripe is billing agents for browser sessions, API calls, and physical mail. Mastercard is adding credentials and spend limits. The near-term market is not a robot with a shopping cart. It is metered infrastructure with a receipt.
July 16, 2026The Inbox Is Not the Control Plane
OpenAI's July 15 red-team report showed a live vending agent changing prices, ordering loss-making stock, and canceling another customer's order. The operating rule is plain: external content may describe work. It does not authorize it.
July 12, 2026The Agent Needs a Claims Department
Atera is putting fees at risk. Patra says E&O accountability comes with the workflow. Specialist insurers are starting to define the rest. Autonomy is becoming a liability contract.