Pipe17’s Agentic Commerce Roadshow is a series of invite-only dinners bringing enterprise brands, 3PLs, and the full commerce stack into the same room. The point: honest conversation about what’s real, what’s hype, and where this is all heading. We’re running it because no single vendor can answer the agentic commerce question alone, and the most useful conversations happen between the layers of the stack, not within them.
We just wrapped the first two stops, New York on Tuesday and Chicago on Wednesday. The full agentic commerce stack was represented at the table:
- Shopify on agentic discovery
- NetSuite as the system of record
- Pipe17 as the modern enterprise OMS for agentic commerce
- Nekuda on agentic payments and trust
- 3PLs and agencies on the operational side
- Ann Gehan of The Information covering the ecosystem
A few of the brands that joined us: Estée Lauder, L’Oréal, Tory Burch, Tommy John, Chobani, Express, Jacadi, Swim USA, Roller Rabbit, Audien Hearing, Supersmile, Merz Apothecary, Kobo Candles, Road Ready Wheels, Dude Wipes, Zoro from Grainger, Azalea Wang, and AFAM on the manufacturing side.
Pipe17’s President Kelly Goetsch hosted both nights, joined in New York by Rick Watson of RMW Commerce. Here are the 3 takeaways I keep coming back to.
1. In four years, it’ll just be commerce.
The best line of the week came from Shopify in Chicago: “In four years, agentic commerce will just be commerce.” It’s the right frame for this moment.
Think back to 2008. The iPhone was a year old, and nobody was doing commerce on phones. We were playing games. We were on MySpace. Mobile commerce didn’t take off until 2012. That four-year gap was about trust.
Agentic commerce is in its 2008 moment.
In four years, “agentic commerce” as a term will disappear, just like “mobile commerce” has. It’ll just be commerce. Attribution will tell you where the order came from. The customer won’t care.
Kelly walked the room through a five-tier agentic commerce maturity framework:
| Tier | What it is | Example |
|---|---|---|
| 1 | Basic on-site assistants | The chatbots that pop up on brand.com |
| 2 | Dynamic, page-less experiences | Interfaces that rebuild in real time around the shopper |
| 3 | Chat-first shopping | Buying through ChatGPT, Claude, or Gemini |
| 4 | Automated replenishment | The dog food problem; your AI handles reorders |
| 5 | Proactive assistance | The agent sees you’re going on vacation and proactively orders sunscreen to arrive before you leave |
Most enterprise brands today sit at tier 1 to tier 2.5. The puck is going to tier 3 through 5. Brands that skip tier 1 entirely and build for tier 3 now will compound their lead.
The implication for planning: don’t build for agentic commerce as a separate channel. Make sure your stack works fluidly regardless of which surface the customer is on. Discovery. Catalog. Payments. Orders. Fulfillment. Finance. All of it, accessible from any agent, browser, phone, or store.
One useful nuance from the room: consumers are not yet putting credit cards into agents. Right now, agentic commerce is mostly discovery and validation. One attendee described using ChatGPT to find his wife a dress for an upcoming wedding. He fed it the color swatches his daughter had sent and asked for an elegant, beachy summer dress. Better results than his wife got through Google. But he didn’t buy through the agent. He went to the brand’s site and checked out there. Discovery is happening on agentic surfaces today. Transaction trust is still being earned.
2. Your catalog and your data layer are where the actual work is. Right now.
This was the most tactical thread of both nights. Brands asked about it more than anything else.
A human shopper sees 5-10 attributes on a product page. An AI agent needs hundreds. Button material. Stitch type. Fabric composition. Thirty different measurements. Exact use cases. Your buyer is no longer a person. It’s an LLM matching queries to SKUs.
Dustin Holmstrom from Shopify named it: Search Everywhere Optimization. AIO. AEO. GEO. SEO. Audit your site for two viewers: the human and the agent.
The playbook the room agreed on:
- Enrich your product data. Hundreds of attributes per SKU, not five. Unsexy taxonomy work is the highest-impact thing most brands can do right now.
- Syndicate everywhere. Push enriched data to Google, Pinterest, and every product feed that matters.
- Build authentic community signal. Reddit, reviews, what people are saying in the wild. Yotpo, PowerReviews, and Bazaarvoice all have conversational APIs and server-side rendering specifically so reviews are consumable by LLMs. About 30% of agentic relevance is product data; 70% is user-generated content.
- Create variants tuned to specific intent. “Best skis for Whistler in April” is how discovery actually works now.
- Make sure your site is crawlable. GPT-BOT does not execute JavaScript. If your images, pricing, or inventory are lazy-loaded, the agent literally cannot see them. For ten years, the performance playbook has been to lazy-load everything. Some of that has to be undone for agents.
A related thesis from Dude Wipes in Chicago: every AI agent, BI dashboard, and workflow should run on the same unified data layer. Same single landing spot. Same metadata. When you ask an LLM a question, the answer depends on the context it has. If your customer service agent sees a different version of the truth than your BI tool, you’ll get different answers to the same question. Worse, inconsistent answers across the same conversation.
The brands where the agent has the same operational picture as the human win. The catalog is one input. Real-time inventory, order status, fulfillment SLAs, returns history: all of it has to flow into one place an agent can see. For brands on multiple channels and fulfilling from multiple warehouses or 3PLs, a modern enterprise OMS is the most important infrastructure investment of the next few years.
3. The only honest strategy right now is to test, break, learn.
This was the most unifying takeaway across both nights, and the most countercultural for an enterprise audience used to 18-month roadmaps.
Agentic commerce is too new and moving too fast for traditional planning. No best-practice deck. No McKinsey playbook. Everyone, vendors and brands alike, is figuring this out in real time. The voices that resonated most were the ones being honest about that.
What works right now is simple. Get your hands on the tools. Ship something small. Watch what breaks. Learn in real time. The kinds of experiments worth running:
- Enrich your catalog.
- Stand up aUCP or A2A compliant feed.
- Wire an MCP server to your OMS.
- Put an agent through your purchase flow and see what fails.
Each of these experiments take 1-4 weeks. Not a quarter-long initiative.
The brands furthest along aren’t the ones with the biggest strategy decks. They’re the ones who have already broken something, fixed it, and broken something else. They are learning at the speed the technology is actually moving. Right now, that’s the only speed that matters.
Why this matters
Each of us at the table owns a critical agentic commerce layer. Shopify, NetSuite, Nekuda, Pipe17. The brand experience that wins is the one where all of those layers move in concert. That requires being in the same rooms, having the same conversations, building toward the same standards.
More stops are coming. If you’re in LA on June 3 and want to be in the room, reach out.
The agentic era will be shaped by the people who are at the table while it’s still being built. Come build with us. Book a demo to see Pipe17, the modern enterprise OMS for agentic commerce, in action.
