Conduit logoConduit

Research

What agents are told, and what agents can do.

We measure the agentic commerce ecosystem two ways: scripted probes that try to complete a real checkout, and manual checks on what storefronts actually declare. The two answers disagree, and the disagreement is the interesting part.

We watch it for B2B, where the buyer is a procurement lead rather than a shopper. Each report is a dated snapshot and ships the numbers behind it as JSON and CSV, so you can check the charts rather than trust them.

Yuri Goncharenko

Yuri Goncharenko

Founder, Conduit

Published 2 September 2026

founder@aiconduit.app

Latest reading

31 August 2026

Verified merchants
15,942
Domains monitored
19,540
AP2 mandate
9
Embedded checkout
1

10 readings on record, 29 June 2026 to 31 August 2026. Source: UCP Checker.

Editions

How this is made

Two halves, labelled separately on every page. The ecosystem readings are UCP Checker’s public data, recorded here by hand. Nothing in our stack computes them and we do not present them as ours.

We measure two ways, and both are ours. Scripted probes run agent clients against live storefronts and attempt a real checkout. They are read-only by design and never send a payment instrument. Manual checks then go store by store on what the scripts surface, because a declared capability and a working one are different things and only one of them survives contact. Each finding carries the date it was run.

We are watching this for B2B. The public buildout is consumer storefronts, and that is not our question. Ours is whether an agent can place a purchase order against approved terms with a distributor. We monitor the gap in order to close it.

Corrections stay visible. Where a reading moves backwards we leave it and say why, rather than smoothing a chart into a story.

Each report is a dated snapshot. We publish when there is something worth publishing, not on a schedule.

Guides and explainers

Background reading on the protocols and the mechanics, separate from the measurements above.

Questions

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.
Which protocols power agentic commerce?
The main open standards are UCP (Universal Commerce Protocol, by Shopify and Google), ACP (Agentic Commerce Protocol, by OpenAI and Stripe), and MCP (Model Context Protocol, by Anthropic) for data and tools, plus AP2 and x402 for agent payments and A2A/A2CN for agent-to-agent negotiation. Conduit speaks all of them so an agent needs only one identity.
When will agentic commerce payments be possible?
They are live now: Conduit’s Agentic Procurement MCP v1 lets AI agents discover and pay across thousands of merchants through a single MCP endpoint. v1 ships with multiple payable rails (x402 web-native crypto, cash-on-delivery (COD), and any card on merchant handoff, where a human completes checkout) within a human-approved mandate, plus a one-click human hand-off wherever a merchant still requires one. Invoice (B2B net terms) is coming soon. Fully agent-held card vault for autonomous card checkout comes later. There is no per-transaction fee between an agent and a merchant.