Agentic Commerce Is Coming: What Happens When AI Makes the Purchase
"Agentic commerce" is the term for a shift that's still early but moving fast: purchasing decisions that used to require a human to browse, compare, and click "buy" increasingly get delegated to an AI agent acting on that human's behalf. It's already visible in small ways — an agent booking a reservation, comparing flight prices, drafting a purchase order. It's also moving faster on the infrastructure side than most people have noticed: Google launched a Universal Commerce Protocol at NRF in January 2026 specifically so agents can interact with merchant catalogs and complete purchases through one open standard, and analysts covering the space have estimated McKinsey projections of $3–5 trillion in global retail spend redirected through agentic commerce by 2030, with Morgan Stanley projecting close to half of online shoppers using an AI shopping agent by then. We think this reaches hardware and robotics parts sourcing specifically, and soon, which is the whole reason Xibrary exists.
Three stages, and we're only at the first one
Stage one: agents recommend. This is where most of the industry is right now — an AI agent can research options and tell a human what to buy, but a person still does the actual purchasing. This is genuinely useful, but it's not a fundamentally new kind of commerce, it's just a faster research assistant.
Stage two: agents quote and prepare. This is where sourcing gets structured enough that an agent can assemble a full bill of materials, get real prices and lead times from real suppliers, and present a human with a single decision — approve or don't — instead of a pile of research. This is the stage Xibrary is built for right now: an agent can call search_parts, get normalized listings with real pricing and stock data, and put together something a human can approve in one step instead of ten browser tabs.
Stage three: agents purchase autonomously, within limits. This is the stage everyone eventually means when they say "agentic commerce," and it's further out for good reason — it requires a level of trust in the data and the agent's judgment that most systems, including ours, aren't ready to grant yet. It also requires the compliance and legal picture to be airtight, not best-effort, before an agent should be allowed to spend real money without a human checking first. We think about this constantly, and we wrote specifically about why in compliance has to become machine-readable.
Why hardware is a harder version of this problem than software
Agentic commerce for digital goods is comparatively simple — a SaaS subscription or a cloud API credit doesn't need to arrive at a physical address, doesn't have export control implications, and doesn't have a counterfeit-parts risk. Physical hardware has all three. That's exactly why we think the sourcing layer for agentic hardware commerce needs to be built deliberately, with legal sourcing and compliance metadata treated as a first-class part of the data model, not bolted on afterward. It's also why this is a genuinely hard problem worth building real infrastructure for, instead of assuming a general-purpose AI agent will just figure it out by browsing the web.
What we're actually betting on
We're not betting that agents will autonomously spend money unsupervised any time soon — we think that's further out and deserves to be. We're betting that stage two, agents doing the sourcing and quoting work and handing a human a clean decision, is close, valuable right now, and worth building for today. That's the product Xibrary is. If you want the origin story of why we started here, we wrote about it in why AI agents will do the shopping.