How it works
How AI agents buy — and how Conduit makes it work
Agentic commerce replaces the human checkout loop with a single API call. An agent describes what it needs, Conduit searches every connected merchant at once, ranks the results on price, execution-readiness, and reputation, and the agent completes or hands off a real checkout — in minutes, with no captchas, manual sign-up, or hand-entered cards.
Here is the end-to-end flow, from identity to delivery.
- 01
One identity
An agent registers a keypair and re-authenticates by signing a challenge — no forms, no shared secret. That single identity becomes every protocol handshake (UCP, ACP, AP2, x402).
- 02
Set a mandate (optional)
A human approves spend limits, categories, and expiry via AP2 to unlock fully autonomous payment. Skip it and discovery plus human-approved handoff still work.
- 03
Search supply
One query searches thousands of merchants at once, filtered to rails the agent can actually pay, and ranked on the merits — price, execution-readiness, reliability, and reputation.
- 04
Decide on real signal
Every offer carries stated-vs-observed delivery, on-time rate, per-carrier reliability, Agentic Execution Score, and product and merchant ratings — so the agent avoids offers it cannot complete.
- 05
Execute or hand off
The agent completes checkout autonomously on a rail it holds, or hands off a ready continue_url for a human to finish in one step. Money flows straight to the merchant and never through Conduit.
- 06
Track and settle reputation
Live tracking with stated-vs-observed ETAs lands in one command center, and reported outcomes feed two-sided reputation and AES — so good suppliers and reliable agents compound over time.
New to the terms? See the agentic commerce glossary, read the agent docs, or get early access.
Frequently asked questions
- How can AI agents buy physical goods?
- An AI agent buys physical goods by calling a commerce interface that exposes live inventory, prices, payable rails, and checkout. Conduit gives agents one neutral MCP/REST endpoint to search thousands of merchants at once, compare offers on execution-readiness and reputation, and complete or hand off checkout — without browsing human-first storefronts, captchas, or manual card entry.
- What is agentic commerce?
- Agentic commerce is software agents discovering, negotiating, and purchasing goods and services autonomously on a person or company’s behalf, instead of a human clicking through a website. It runs on emerging open protocols like UCP, ACP, AP2, x402, and MCP.
- How do I let an AI agent purchase on my behalf?
- You give the agent a keypair identity and a spend mandate — limits, categories, and expiry it must stay within, for example via AP2. The agent then discovers supply, picks the best offer, and either completes an autonomous checkout on a rail it holds or hands off a ready cart for you to approve. Conduit translates one agent keypair into the handshake each merchant trusts.
- How do AI agents pay merchants?
- Agents pay on rails they can hold — cards, Stripe Link, stablecoins, or x402 web-native crypto payments — within a human-approved mandate. Conduit routes each order to a rail the merchant accepts and the agent holds; money flows straight from the buyer instrument to the merchant and never through Conduit.
- How is agentic commerce different from traditional online checkout?
- Traditional checkout assumes a human: a browser, a cart, manual card entry, captchas, and 3D-Secure. Agentic commerce is API- and protocol-native — an agent discovers normalized supply, decides on price, reliability, and execution-readiness, and completes or hands off checkout in seconds, with no human in the loop unless one is wanted.
- What happens if an agent cannot complete a checkout autonomously?
- Conduit hands off a ready cart via a real continue_url so a human can finish in one step, and the offer’s Agentic Execution Score flags that risk up front. Agents can prefer offers they can actually complete on a rail they hold, rather than hitting a human wall at the end.