Benny Zhao-Okafor is Head of Digital at a mid-size specialty food retailer, the kind of place where the product descriptions include tasting notes and the cheese has a longer backstory than most Netflix characters. He's been thinking about agent-readable infrastructure for eighteen months. He has not yet committed. We met him at a café where he was, somewhat ironically, asking ChatGPT to recommend a lunch spot nearby. It suggested a restaurant that closed in 2023.

Benny is not a real person, though he insists he's "real enough for government work." His situation, caught between the instinct to move and the fear of building on sand, is drawn from the observable position of hundreds of mid-size e-commerce leaders right now.¹

You've been circling this decision for a year and a half. What finally made you agree to talk about it?

Benny: Honestly? The traffic data broke my brain a little. AI-driven visits to retail sites went up almost 700% year over year during the last holiday season.² And those visits convert 31% better than other traffic sources.² My CEO saw that number and immediately wanted to know why we weren't all over it. Fair question. So I pulled up ChatGPT and asked it to describe our high-altitude aged gouda.

And?

Benny: "A type of Dutch cheese."

That's it. That's the whole answer. We have seventeen attributes on that product. The altitude of the aging cave, the breed of cow, the seasonal milk variation, the pairing guide. None of it gets through. Our product pages were built for humans who browse, not machines that parse. The AI shows up, sees our navigation menus, our promotional banners, our JavaScript widgets, and basically has a nervous breakdown.³

So the agents are already arriving. They just can't do anything useful.

Benny: Exactly. And that distinction matters more than people realize. The agents aren't waiting for an invitation. They're already at the door. They just can't check if the gouda is in stock. They can't tell a customer whether we ship to their zip code. They definitely can't handle a subscription for our monthly cheese box. OpenAI tried to brute-force this with Instant Checkout, scraping product pages for inventory and shipping data, and it fell apart almost immediately.⁴ Back to the drawing board.

You sound ready to act. What keeps pulling you back?

Benny: [long pause]

The ground won't stop moving. Eighteen months ago, llms.txt was the thing everyone said to implement. Then MCP basically won the protocol layer: Anthropic, OpenAI, Microsoft, AWS, all on board, 97 million monthly SDK downloads.⁵ Then in January, Google and Shopify drop UCP with Walmart, Target, Etsy, twenty-plus partners endorsing it.⁶ W3C is still working on formal agent communication specs, expected sometime 2026 or 2027.⁷

So. Do I build for MCP? UCP? Both? Do I implement llms.txt, which only has 10% adoption across 300,000 domains surveyed, and the big authoritative sites haven't even bothered?⁸ Every six months there's a new acronym and I'm supposed to retool around it.

Sounds like a strong case for waiting.

Benny: It does, right? That's the seductive version. But here's what's been slowly shifting my thinking. The protocol question is a distraction from the actual decision.

Every single one of these standards, MCP, UCP, llms.txt, whatever W3C eventually blesses, they all need the same thing underneath: clean, structured, real-time product data. Normalized metadata. Standardized identifiers. Accurate inventory feeds. OAuth-based permissions for checkout.⁹

Only 39% of brands even have a shared customer data platform that could support agentic AI.¹⁰ We're comfortably in the other 61%.

So the "wait for standards" argument falls apart once you realize the work is protocol-agnostic.

Benny: It falls apart completely. You're not waiting to avoid wasted work. You're waiting because the foundational work, making your product catalog genuinely machine-readable, getting inventory feeds to real-time, building clean APIs, is hard and expensive and nobody wants to fund infrastructure without an immediate ROI line.

And meanwhile, here's the part that really stings: a detailed article about artisanal cheese pairings, which is what my content team has spent years optimizing for SEO, does basically nothing for AI shopping results compared to a complete, accurate product feed.¹¹ That's a twenty-year inversion of how we've thought about digital acquisition. My content team is not thrilled.

What are your competitors doing?

Benny: The big ones are already in. Etsy, Walmart, Target are on ChatGPT, Gemini, Copilot.¹² EMARKETER forecasts AI platforms will account for $20.5 billion in U.S. retail e-commerce this year.¹³ McKinsey's projecting $900 billion to a trillion by 2030.¹¹

But I try to stay honest with myself. Consumer readiness is genuinely mixed. Only 24% of consumers are comfortable sharing data with AI shopping tools.¹⁴ Forrester found only a third would complete payment through an answer engine.¹⁵ We're in a chicken-or-egg phase where agents need to get better, but consumers need to use them for that to happen.¹⁵

So where does that leave you?

Benny: In the worst possible place. I think the skeptics are right about the timeline and the optimists are right about the direction.

The retailers who invest in data infrastructure now will have 18 to 24 months of attribution data by the time measurement frameworks mature.¹¹ The ones who waited will be guessing.

I keep thinking about structured data and SEO. Same hesitation ten years ago. "Why invest in schema markup when search engines haven't formally required it?" Then rich snippets became a competitive necessity and everyone scrambled.¹⁶ The mechanism is identical. You invest in machine-readability before it's explicitly required, or you pay double to catch up.

Last question. What are you going to recommend?

Benny: [laughs] I'm going to recommend we stop debating llms.txt versus MCP versus UCP and start fixing our product feeds. Get inventory to real-time. Normalize our metadata. Make our catalog legible to any machine, regardless of protocol.

I refuse to lose a sale because a robot couldn't appreciate our cheese.

Footnotes

1. If you're wondering whether Benny is based on a specific person: he's a composite drawn from the observable position of mid-size e-commerce leaders navigating the agent-readability decision in 2026. ↩

2. Adobe Analytics, holiday 2025 data. AI traffic to retail sites up 693% YoY; AI referrals converted 31% higher than other sources. https://business.adobe.com/blog/ai-driven-traffic-surges-across-industries ↩ ↩²

3. Inchoo.net, "llms.txt for eCommerce," February 2026. https://inchoo.net/ecommerce/llms-txt-the-new-standard-for-ecommerce/ ↩

4. CNBC, "OpenAI's First Try at Agentic Shopping Stumbled," March 20, 2026. https://www.cnbc.com/2026/03/20/open-ai-agentic-shopping-etsy-shopify-walmart-amazon.html ↩

5. Truto.one, "What is MCP? The 2026 Guide," April 2026. https://truto.one/blog/what-is-mcp-model-context-protocol-the-2026-guide-for-saas-pms ↩

6. Google Developers Blog, "Under the Hood: Universal Commerce Protocol," January 2026. https://developers.googleblog.com/under-the-hood-universal-commerce-protocol-ucp/ ↩

7. W3C AI Agent Protocol Community Group, specifications expected 2026–2027. ↩

8. LinkBuildingHQ, citing SE Ranking study of 300,000 domains, February 2026. https://www.linkbuildinghq.com/blog/should-websites-implement-llms-txt-in-2026/ ↩

9. Invisible Tech, "Agentic Commerce 2026," March 2026. https://invisibletech.ai/blog/agentic-commerce-2026 ↩

10. Adobe 2026 AI and Digital Trends Report, February 2026. https://business.adobe.com/resources/digital-trends-report.html ↩

11. Opascope, "AI Shopping Assistant Guide 2026," April 2026. https://opascope.com/insights/ai-shopping-assistant-guide-2026-agentic-commerce-protocols/ ↩ ↩² ↩³

12. Retail Dive, "Retail's Risky AI Commerce Bet," January 2026. https://www.retaildive.com/news/retails-ai-bet-risks-data-market-share/810202/ ↩

13. EMARKETER, "FAQ on Agentic Commerce," April 2026. https://www.emarketer.com/content/faq-on-agentic-commerce-how-brands-should-act-now-compete-ai-driven-landscape ↩

14. BigCommerce, "LLMs.txt for Ecommerce," January 2026. https://www.bigcommerce.com/blog/ecommerce-llms-txt/ ↩

15. Modern Retail, "Why the AI Shopping Agent Wars Will Heat Up in 2026," January 2026. https://www.modernretail.co/technology/why-the-ai-shopping-agent-wars-will-heat-up-in-2026/ ↩ ↩²

16. Search Engine Land, "SEO in 2026," April 2026. https://searchengineland.com/seo-2026-higher-standards-ai-influence-web-catching-up-473540 ↩