An Interview With Eric Seufert About Agentic Commerce (Stratechery, 2026-11-19)
Why this is in the vault
Seufert (ad analyst, MDM.co) argues agentic commerce resolves into advertising rather than take rates, while Thompson pushes back from firsthand experience that consumers will favor retailers who welcome third-party agents; the disagreement is the useful part.
Mapping against Ray Data Co
Concrete connection: Thompson's bolt-buying anecdote is a field report on what agents need from a seller's site - he uses Claude Code to plan a project and fill carts, and he stopped visiting small retailers that blocked agents while rewarding the one that let the agent fill the cart. That bears directly on RDCO's public surfaces (Scribble Works storefront, Sanity Check, HQ pages): if agents are the discovery path, an agent-readable, non-hostile site is a distribution decision, and Scribble Works' "parents site, kids paper" principle should be checked against whether an agent can browse and assemble a purchase without an account wall. Second, Seufert's framing that third-party agents are commodity interfaces whose only durable monetization is ads sharpens the credibility-for-phData angle: client conversations about "agentic commerce" in retail/CPG can be anchored on his hard/soft/discovery taxonomy and the Walmart "hand the request to Sparky" API pattern rather than vendor hype. Mapping is medium: strong framing material, no immediate build or decision.
The core argument
- Seufert's thesis (from "Agentic Commerce Is a Mirage," Sept 2025): Amazon blocks third-party agents because disintermediation kills its high-margin ad revenue. The resolution is ads, not agent take rates. He treats the OpenAI-Amazon ads deal as the first prototype (Amazon DSP demand into ChatGPT: consumer, ChatGPT and Amazon all win) and Google's Universal Cart as the second (price tracking and carts across retailers, monetized with Direct Offers ads; retailers choose checkout on-site vs in-Google).
- Amazon as identity layer: he claims deterministic identity on roughly 90% of US households lets Amazon inject demand into Roku, Disney+, Netflix, Tubi, OpenAI and Pinterest without cookies. Meta-Amazon partnership appears to have fizzled; the idea persists.
- Muse: Meta's promise not to link Muse data to Meta data is largely moot because purchases flow back through the Conversions API anyway. Both argue Meta should simply say it will monetize with ads.
- Taxonomy: hard agentic commerce (agent buys autonomously), soft (agent shortlists, user picks, agent checks out), agentic discovery (agent finds, user checks out). Amazon wants discovery only via its own ads; Shopify and Walmart gate hard/soft behind human approval or their own agent. Walmart's agentic principles route outside agents to Sparky, so Sparky's output (with ads) mediates everything.
- Take rates collapse: Shopify's CEO reportedly called 2% unrealistic; OpenAI's Instant Checkout tried 4%. With transaction fees near basis points, discovery monetizes via ads. Seufert says trust fears about ads in agent output are a design problem (clear demarcation, e.g. sponsored chats).
- Where they diverge: Thompson argues convenience lies in cart construction inside his own agent workflow, not checkout, so Alexa-for-shopping texting does not beat his agent; Seufert concedes the point for workflow-embedded purchases but holds that Amazon's leverage (fulfillment speed and cost) outweighs Muse's. Both agree small retailers blocking agents (Cloudflare-style default hostility) hurt themselves and that being agent-discoverable now may produce sticky preference.
- Google and ChatGPT: Google's lower-monetizing AI Mode is framed as a replay of Stories/Reels (more inventory, lower price per ad, more engagement). Seufert's multi-turn ad funnel: brand ads early, higher-bid conversion ads as context accrues, multiple ads per conversation. He estimates ChatGPT ads near a $1B run rate and moving toward the $2.5B target; Netflix missed its roughly $3B ad target (about $2.7-2.8B) and was slow on pixel/CAPI infrastructure.
⚠️ Sponsorship
Issue is "Presented by Stripe" (Stratechery's first paid sponsor, disclosed 2026-10-21; see [[2026-10-21-stratechery-presented-by-stripe-sponsorship]]). The mid-issue ad block is Stripe-owned OpenRouter (median customer model families rising from three to six). Bias implication: Stripe appears inside the discussion as a "pipe" with Link vs Shop Pay frenemy framing and the OpenRouter acquisition; per Thompson's own disclosure, treat Stripe/fintech commentary as commercially interested. Separately, Seufert sells MDM.co subscriptions and has a long-running public position ("Mirage" series, now part 5) that this interview largely confirms; Thompson plugs it and a retailer (Albany County Fasteners) unpaid.
Related
- [[2026-05-28-stratechery-eric-seufert-interview-models-ads-ai-upside]]
- [[2026-09-22-stratechery-amazon-blocks-muse-aggregator-moat]]
- [[2026-10-01-stratechery-delrey-interview-amazon-walmart-muse]]
- [[2026-09-15-stratechery-openai-amazon-ads-walmart-apple-pay]]
- [[2026-10-21-stratechery-presented-by-stripe-sponsorship]]