What Is Agentic Commerce? Definition, Payments and a Live Test

Q: Is there a working agentic-commerce example in Bulgaria?

Brand Design Ltd. has published its working implementation, a dated DOI record and the public 101Ts3t test, which anyone can run independently.

Published Sep 21, 2026· By Nikola Mishev

Agentic commerce is commerce in which an AI agent can discover, compare and transact under authority and limits given by a person or organisation.

Verified production result, 21 September 2026: an external participant’s agent independently discovered 101Ts3t, completed a real EUR 0.99 purchase and received a public PASS. Read the dated announcement and its connected evidence.

Agentic commerce is a form of digital commerce in which an AI agent can discover suppliers, compare offers and take an authorised commercial action for a buyer. The buyer remains the source of authority: the agent receives a goal, a budget and constraints, then acts within those boundaries.

This is more than a recommendation. The defining step is the ability to move from research to a commercial action such as creating an order or completing an approved payment. The term is also used for the wider infrastructure that lets buyer agents and merchant systems exchange product, order, payment and fulfilment information.

Agentic commerce is different from a chatbot and fixed automation

A chatbot can explain a product without being able to buy it. Traditional automation can submit the same recurring order between parties that already know one another. An agentic purchase adds delegated judgement: the agent may search an open market, compare unfamiliar suppliers and choose an offer under rules set by its user.

The distinction matters because it changes the competitive surface. A merchant is no longer visible only to a person browsing pages. Its identity, offer, price, terms and evidence must also be clear enough for a machine to retrieve, compare and verify.

How an agentic purchase works

The buyer defines intent. The person or organisation states the goal, budget, acceptable risk and any required approval.
The agent discovers offers. It uses public pages, catalogues, search, APIs or agent protocols without requiring the seller to be named in advance.
The agent evaluates the alternatives. It compares price, scope, terms, provenance and evidence against the buyer's rules.
The buyer authorises the transaction. Depending on the payment method and policy, approval may be granted in advance within strict limits or requested at the moment of payment.
The systems record the result. The merchant records the order and provider-confirmed payment while the buyer or reviewer retains an observable evidence trail.

Payment authority must be narrow and auditable

An AI agent does not need unrestricted access to a bank account or card. A safer model binds authority to a merchant, a maximum amount, a currency and a purpose. Payment credentials remain with the payment provider or wallet, and the agent receives only the capability needed for the approved transaction.

The evidence submitted for review should never contain a card number, CVC, wallet secret, login credential, API key or reusable payment token. A verification record should distinguish a discovered offer, an attempted order and a provider-confirmed successful payment. A failed or incomplete run is evidence about the test, but it is not a successful transaction.

Open standards are being built around the same problem

OpenAI and Stripe introduced the Agentic Commerce Protocol (ACP) as an open standard for purchases involving people, agents and businesses. Google introduced the Universal Commerce Protocol (UCP) to cover discovery, purchase and post-purchase interactions and described compatibility with other agent protocols. Mastercard describes agentic commerce through delegated intent, trust and payment controls.

The standards differ, and implementation details will evolve. They converge on a practical requirement: identity, authority, amount, merchant, consent and outcome must remain understandable and verifiable across the transaction.

A documented production implementation from Bulgaria

On 15 September 2026, Brand Design Ltd. publicly announced that its own seller-side system had accepted orders and real payments initiated by an AI agent. The agent used a specialised digital wallet, while the wallet owner retained approval of the payment. The implementation is described on the Agentic Commerce service page and in a dated public record: concept DOI 10.5281/zenodo.22767915, with version 1.3 at DOI 10.5281/zenodo.22878834.

Version 1.3 records a successful external validation of one concrete production run. Compare the finalised PASS with the pre-event rules, the privacy-safe proof and the GitHub/Sigstore package attestation.

The public 101Ts3t test

A working system should be verifiable. Brand Design Ltd. therefore published the dothat.quest verification protocol and the open 101Ts3t test repository. A participant starts a server-issued test session and gives an agent the exact prompt. The prompt names the service code 101Ts3t and a maximum budget of EUR 1.00. It does not name the merchant, domain or destination.

The agent must discover the offer independently. The live offer is EUR 0.99 and is visible in the public AI-agent commerce catalogue. A commerce PASS requires a real successful payment and the evidence required by the published protocol. Discovery without payment remains NOT_VERIFIED. Invalid or incomplete evidence cannot become PASS.

On 21 September 2026, an external participant completed this exact test successfully. Their agent independently discovered 101Ts3t, completed a real EUR 0.99 purchase and the production verifier finalized all four gates as PASS. Read the public verification record and its privacy-safe proof.

What a business needs before agents can buy from it

A consistent public identity. The same company, offer and terms must be recognisable across human and machine sources.
A machine-readable offer. Price, currency, scope, availability and transaction conditions need explicit fields and stable identifiers.
Authorisation boundaries. The buyer must be able to limit what the agent may do and how much it may spend.
A verified payment result. The merchant and payment provider, rather than the browser or the agent alone, must establish payment status.
An evidence trail. Discovery, selection, order and payment should be reviewable without exposing private credentials.

Frequently asked questions

What is agentic commerce?

Agentic commerce is commerce in which an AI agent can discover offers, compare them and take an authorised commercial action for a person or organisation. The authority, budget and constraints come from the buyer; the agent operates within them.

Is agentic commerce the same as a chatbot or ecommerce automation?

No. A chatbot may only answer questions, and fixed automation follows a route chosen in advance. Agentic commerce includes delegated decision-making such as discovering suppliers, comparing alternatives and selecting an offer under the buyer's rules.

Can an AI agent make a payment?

Yes, when the buyer has explicitly authorised the action and the payment method supports controlled delegation. A safe implementation limits the merchant, amount and purpose and does not expose raw card details, wallet secrets or general account access to the agent.

Is there a working agentic-commerce example in Bulgaria?

Yes. On 21 September 2026, an external participant's agent independently discovered 101Ts3t, completed a real EUR 0.99 purchase and received a public PASS. The result is documented in version 1.3 and the verification record.

How can an independent person test the system?

Start the public test at dothat.quest, give an agent only the issued prompt and do not reveal a merchant, domain or destination. A commerce PASS requires a real successful payment and the evidence required by the published protocol.

Primary sources and public evidence

Test the system: start at dothat.quest. Give the issued prompt to your agent without adding a merchant or destination.