Source: raw/newsletter-zyppy-signal-1c86d2900d.md — Cyrus Shepard (Signal by Zyppy) interviewing Tim Resnik (ex-Product/SEO lead at Walmart & Wayfair; advised top e-commerce retailers at Botify). Published 2026-04-15. 2026-09-29 addition: raw/Agent_Wars.md and raw/The_Real_Risks_of_AI_Agents.md (The AI Daily Brief, NLW). 2026-10-27 addition: raw/The_9_Stages_of_AI_Agent_Adoption_Explained.md (Greg Isenberg), raw/Ep._244_...Agent-to-Agent_Commerce.md and raw/Ep._247_...MAICON_2026_Recap...md (The Artificial Intelligence Show), raw/How_A.I._Agents_Are_Going_to_Transform_Our_Lives_Hard_Fork.md (Hard Fork).

A practitioner interview on how AI is reshaping e-commerce discovery and what merchants should do now. Evidence class: expert opinion + guidance, no data — and explicitly forward-looking/speculative on agentic-commerce timing. Useful as a prioritization checklist from a credible operator, not as evidence. Distinct from the AI-citation research cluster.

Key Takeaways

  • Product feeds & catalog enrichment are Job #1. Feeds — historically punted to the paid-ads team or an agency — are becoming a primary way AI understands your catalog. AI needs real-time availability and pricing faster than a crawler can fetch it, so the feed becomes the source of truth. Google is building new protocols (a Universal Commerce Protocol, plus Conversation Commerce feed attributes in limited beta) on top of the core feed, surfaced through Google Merchant Center. Action: treat feeds as canonical assets and eliminate errors (most Merchant Center feeds are “riddled with errors”).
  • SEO basics matter more, not less, in an AI world. Non-Google AI bots — “especially anything coming from OpenAI” — struggle with heavy JavaScript. Slow load times, poor architecture, and JS-dependence can stop AI bots from seeing products at all. Google “would prefer really fast HTML” (crawling JS is expensive; Google is spending on TPUs/data centers). Action: fast load, good internal linking, minimal reliance on hard-to-parse JS before any new AI tactic.
  • Brand & loyalty play outsize roles in “perfect information” markets. When AI platforms hold millions of data points on you and every competitor, price competition gets harder and on-platform purchases erode your advantages further. Brand affinity and loyalty (“sometimes irrational, but still loyalty you can build on”) become the differentiator. Action: shift budget from head keywords (e.g. “best mattress online”) toward your brand (“Mattress Kingdom”); invest in digital PR that showcases brand + product.
  • True agentic commerce isn’t here yet. Protocols are rolling out, but almost all purchases still go through traditional routes. Full “delegated autonomy” (your agent restocks orange juice when the fridge is empty, end-to-end) is “not even close” per Resnik. Don’t chase shiny objects; get ready.
  • The likely winners: companies with operational excellence, that nail SEO fundamentals tighter than everyone else, and that have a strong brand presence.

Why It Matters

This is the e-commerce cut of the same thesis running through Shepard’s other 2026 pieces: as AI intermediates discovery, defensible advantages move to owned data (feeds), technical accessibility (fast HTML), and brand — not keyword volume. It operationalizes “Transaction Pages” and “Strong Brand” from the winning-content analysis for product businesses specifically.

Agent Wars — platforms pick sides (2026-09-29 addition)

Sources: raw/Agent_Wars.md (The AI Daily Brief, NLW, youtube.com/watch?v=WEqCpG206sc) and raw/The_Real_Risks_of_AI_Agents.md (same show, youtube.com/watch?v=fXRagdr-Q34). The host relays news and third-party posts; none of it is first-party. Auto-caption transcripts.

Five months after Resnik’s interview, a consumer agent that shops became a breakout success, and the platforms split over it. Meta’s Muse, a personal agent, passed ChatGPT to become the number one free app in the US.

  • Amazon blocked Muse.
    • The block. “Effective immediately, Muse agents won’t be able to shop for Amazon products on behalf of their users.” The on-site pop-up: “Continued access by an unauthorized AI agent violates Amazon’s conditions of use.”
    • Amazon’s statement. Third-party apps that buy on customers’ behalf “should operate openly and respect service provider decisions about whether or not to participate.”
    • Context. NLW notes that Amazon runs its own shopping agent and sued Perplexity in November for getting around its agent blockers.
    • One reading. YouTuber Joseph Carlson points to the ad money: Amazon made $76B from advertising in the last 12 months, and “Agents don’t look at ads.”
  • Shopify took the other side. It announced a Meta partnership: “Similar to their partnership with OpenAI, Shopify will allow Muse to access their back end directly.” Muse also gets official support in Shop Pay, Shopify’s agentic checkout, “across all Shopify stores.”
  • Card networks are in. Mastercard joined Visa in officially supporting virtual cards for agents. Mastercard CPO Jorn Lambert: “it’s not about if, it’s about when, and how quickly.”
  • “Everything becomes B2A.” Nicolas Bustamante says Muse bought him socks, ordered groceries, booked a cleaning service and got him a burger for dinner, all within 24 hours. His conclusion: “every business will be selling not just to humans but to their agents.” Box’s Aaron Levie added: “The monetization potential of personal agents that are transacting on your behalf is quite significant.”
  • A question for ad budgets. On “The Real Risks of AI Agents,” NLW asks: “if 10 or 20% of consumer purchases are now mediated by AIs that don’t care about digital advertising,” is that enough to disrupt the ad industry? He poses it as a question, not a finding.

What this changes in the advice above:

  • Feeds become back-end access. Resnik’s “feeds are Job #1” now extends past crawlers and Merchant Center. Agents reach a catalog through platform back-ends (Shopify-to-Muse, Shopify-to-OpenAI). Where you sell decides which agents can buy from you.
  • Resnik’s timing call is not contradicted. “True agentic commerce isn’t here yet” still holds on volume, because no purchase-volume data is given. What changed is that platforms are now actively granting or blocking agent access. This is the permission problem described in the agent-commerce ToS ceiling.

October 2026 update: three camps, and B2B goes agent-to-agent

The blocking map widens (Greg Isenberg, “The 9 Stages of AI Agent Adoption”, raw/The_9_Stages_of_AI_Agent_Adoption_Explained.md; his summary of public reports). He sorts companies into three camps.

  • Block.
    • Amazon blocks Muse.
    • Delta says it “works with zero outside agents” for shopping or booking and treats them as “unauthorized automated activity”.
    • Yelp lets agents in only if they pay through its data-licensing programme.
  • Block by accident.
    • Zillow says agents acting for real customers are welcome, but its anti-scraping tools sometimes block them anyway.
    • Walmart, a Muse partner, had customers report that Muse kept failing on its site; Walmart said that was an accident.
    • Isenberg’s point: many blocks are security systems that “are just not able to tell the difference between the scraper and your personal assistant”.
  • Welcome.
    • Meta is working with Walmart, Stripe and others on an open standard for how agents and businesses talk to each other.
    • Shopify built a way for merchants to sell through AI channels.
    • Visa and Mastercard are building ways for stores to recognize trusted agents, and are working with Cloudflare on it.
  • His pattern. “The companies with the biggest stores are blocking. The companies with the biggest audiences are just opening the door. And the payment and security companies want to be the toll booth no matter who wins.”

B2B becomes A2A (agent-to-agent) (The Artificial Intelligence Show, Ep. 244, interview with Alibaba.com’s president, named “Quo Jang” in captions). This is a first-party view from a marketplace operator.

  • Matching instead of ranking. Search ranks results so a buyer “can only talk to dozens” of hundreds of thousands of suppliers. In an agent-to-agent network, buyer and supplier agents match across the whole network on requirement, budget and timing.
  • Suppliers who give agents real data get matched. Per the interview, suppliers that provide “their real business data” and “real kind of manufacturer capabilities” find it “more easier for them to connect to this network”.
  • The site becomes part of the agent’s harness. Alibaba.com is being rebuilt so “it’s not only designed for human but it’s also designed for agent as well”, with an “AI mode” that lets users treat the site as their agent. Its agent product (spelled “Axio” in the captions) ships a new version daily, against the main site’s weekly or monthly cadence.

Discovery signal: an event found through ChatGPT (Ep. 247, MAICON 2026 recap). The hosts report that a notable share of attendees found the conference through AI. One international attendee who travelled about 30 hours said they “just asked ChatGPT for like what are the best AI conferences in America”. The event’s app was deliberately a web app, not a phone app, which made it “legible to agents”. This is an anecdote, not measured data.

The reversal in one line (Hard Fork, a host relaying an investor): “we spent 20 years trying to prevent bots from being able to transact on the internet. And now we’re doing the opposite.”

What this adds to the advice above:

  • An accidental block costs the same as a deliberate one. Check whether your bot defenses (CDN, WAF, anti-scraping) challenge agent traffic you would want, and follow the trusted-agent work by Visa, Mastercard and Cloudflare (see Cloudflare Pay-Per-Crawl).
  • For B2B sellers, complete and accurate capability data (specs, capacity, lead times) is what an agent-to-agent marketplace matches on. It is the B2B counterpart of “feeds are Job #1”.

Try It

  1. Audit your Google Merchant Center feed for errors and treat it as a canonical, real-time asset (availability + pricing), not a paid-ads afterthought.
  2. Run an AI-bot accessibility check — confirm important product content renders in raw HTML (server-side), not client-side JS, so OpenAI/Perplexity crawlers can read it.
  3. Shift a slice of budget from head keywords to brand + digital PR to build the loyalty that survives perfect-information AI comparison.
  4. Don’t over-invest in agentic-commerce protocols yet — prepare feeds and fundamentals; the autonomous-purchase future is not imminent.
  5. Check which agents can reach your catalog (2026-09-29). If you sell through Shopify, confirm whether your store is exposed to Muse and OpenAI through Shopify’s partnerships and Shop Pay agentic checkout. If you depend on Amazon, assume third-party agents such as Muse are blocked there.
  6. Test your own bot defenses against a legitimate agent (2026-10-27). Run a consumer agent through your checkout or booking flow and confirm your anti-scraping layer doesn’t block it by accident, as happened at Walmart and Zillow.

Open Questions

  • When does agentic commerce cross from beta to material volume? Resnik says “far off”; no timeline or data offered. As of September 2026 there is still no volume data. Muse’s first-week anecdotes and Amazon’s block are the first signs of platforms fighting over agent access (see Agent Wars above).
  • How do the Universal Commerce Protocol and Conversation Commerce attributes actually work at scale? Both were limited-beta at publication.