The Agent Will See You Now: Agentic Commerce, A Structural Shift

The End of the Search Box

For two decades, e-commerce has been organized around a simple primitive: the search box. Type, scroll, compare, click, buy. Every layer of the $6 trillion global e-commerce stack—from Google Ads to Amazon’s A9 algorithm to Shopify’s storefront templates—was architected around this human-driven, click-intensive paradigm. That paradigm is ending.

AI agents are no longer a research project. They are buying things on your behalf, right now. Amazon’s Rufus has 215 million users and a 55–60% conversion rate among those who engage with it. Every major brand is scrambling to make their product catalogs machine-readable—because in a world where agents do the shopping, being invisible to an algorithm isn’t a marketing problem. It’s an existential one.

The New Stack: From Cost-Per-Click to Cost-Per-Context

The agentic commerce stack looks nothing like the performance marketing stack of the last decade.

Discovery is migrating from keyword search to LLM-mediated conversation: queries like “water bottle that keeps coffee hot for 12 hours and fits in a Subaru cup holder” are happening in ChatGPT, not Google. Traditional SEO is necessary but insufficient. What matters now is Answer Engine Optimization (AEO)—third-party validation, structured data, and brand knowledge bases—and Generative Engine Optimization (GEO), which relies on entity building, topical authority, and credibility signals. Companies helping here include Profound, AirOps, DayDream etc.

Selection is shifting from 10 blue links to 4–6 agent-curated options. The agent doesn’t show ads, it shows answers. If your product data lacks temperature ratings, compatibility details, allergen info, and discount status, you don’t exist in the consideration set.

Transaction is where the protocol wars get interesting, and where the real infrastructure investment thesis lives. Key players include Stripe, Skyfire, Crossmint, Nekuda etc.

🧭 What Does This Mean for Founders/Operators?

• Treat product data completeness as a growth lever. Merchants with 95%+ attribute fill rates see 3–4x higher AI visibility; below 80%, agents skip them entirely. 

• Export your catalog, audit coverage across all attributes, implement JSON-LD schema on every product page, and align PR, SEO, and affiliate campaigns to reinforce the same entity signals simultaneously. 

• Make trust machine-readable: ratings, return policies, and fulfillment signals need to be structured, not buried in your UI.

The Protocol Wars: ACP, MCP, and the Fight for Commerce’s TCP/IP

Three protocols are competing to become the standard communication layer between AI agents and commerce infrastructure:

• Stripe’s ACP captures last-mile checkout. It’s live, Walmart has adopted it, and OpenAI charges 4% on top of it for agent-facilitated Shopify purchases. 
• Anthropic’s MCP operates at the discovery layer, enabling Shopify brand servers to be discoverable by external agents—the key unlock for Shopify becoming a marketplace without building one. 
• The Universal Commerce Protocol (UCP), with founding members Google, Shopify, Etsy, Target, and Wayfair, would allow agents to orchestrate purchases across platforms simultaneously. It’s 12–24 months from meaningful adoption, but potentially the largest infrastructure opportunity if it wins.

🧭 What Does This Mean for Founders/Operators?

• Expect a multi-protocol world—like payments today (Visa, Mastercard, ACH, stablecoin) or application-layer internet standards (SMTP, HTTP, OAuth). 

• The interoperability layer is where the next generation of commerce infrastructure companies will be built. 

• UCP’s momentum is growing: its REST/MCP/A2A-compatible architecture handles mid-transaction cart management, layered loyalty logic, split fulfillment, and reverse logistics—breadth that ACP, a checkout-first spec, is still evolving to match.

Amazon: An Agentic Commerce Test Case

Amazon is the most interesting strategic case because it has the most to lose. Half of all product searches still start on Amazon, but the other half is rapidly migrating to AI-driven discovery. The threat: an OpenAI agent routes purchases to a Shopify store where brands offer lower prices by avoiding Amazon’s 15% take rate. Amazon knows this—over the holidays, it briefly tested directing customers to brand websites, violating one of its most sacred rules. It is war-gaming a world where it becomes the fulfillment API, not the storefront. In April 2026, Amazon joined the UCP Tech Council alongside Meta, Microsoft, Salesforce, and Stripe—securing a governance seat to shape the standard rather than watch it develop without them.

🧭 What Does This Mean for Founders/Operators?

• Model the margin delta before agents do it for you. 

• Amazon’s total take rate often exceeds 30% including FBA—a 10–15% DTC price advantage is enough for price-optimizing agents to route around it. 

• Get on the protocols agents read (UCP, ACP, Google Shopping Graph) like you got on Google Shopping feeds in 2012. 

• Build agent-callable, not just human-browsable. 

• Audit your Amazon price parity clauses—MFN enforcement will intensify as agent-driven price comparison becomes automatic.

The Attribution Black Hole

Here’s the most underappreciated problem in agentic commerce: we can’t measure it. UTM tracking captures less than 10% of AI-influenced purchases. Consumers see a recommendation in ChatGPT, don’t click, Google the brand, and buy on the website—GA4 attributes it to direct or branded search. The AI influence is invisible.

The current best signal: post-purchase surveys asking “How did you discover us?” with ChatGPT and Perplexity as explicit options, yielding 60–80% consumer self-report rates. It’s a band-aid on a structural problem. The company that solves independent AI attribution without relying on walled-garden data from OpenAI or Google will build an extremely valuable business. This is a wide-open investment opportunity.

Brand as Truth Hub: The New Moat

In the agentic world, brands are squeezed between AI owning discovery and marketplaces owning transactions. The only defensible position is owning the context layer—being the authoritative source that dictates how agents understand, evaluate, and recommend your products. Companies solving this include Spangle, FERMÀT, Catalog etc. This means: exhaustive machine-readable catalogs with deep use-case attributes, real-time inventory APIs, structured schema data, third-party validation that builds LLM entity trust, and supervisor agents that review brand representation before a purchase executes. Brands that don’t build this infrastructure won’t lose a bidding war—they’ll simply stop being surfaced.

🧭 What Does This Mean for Founders/Operators?

• Conduct a brand knowledge audit and treat the gaps as your content roadmap. 

• Create answer-complete content (comparison tables, use-case guides, FAQs) that answers not just what your product is, but who it’s for and what makes it different.

The Settlement Layer: Why Payments Infrastructure Matters

Agentic commerce changes settlement. If an agent executes hundreds of micro-transactions daily across merchants and geographies, legacy card rails (3-day settlement, 3.2% acceptance cost, $250B annual fraud) become a bottleneck. Stablecoin settlement offers near-instant finality, pre-funding reserve compression from 3 days to ~0.25 days, and cross-border corridors under 30 seconds. Stripe already offers 1.5% stablecoin take rates vs. 3.2% for cards. The catch: consumer protection. Card networks have decades of dispute resolution infrastructure. Stablecoin rails have none. The company that builds dispute resolution for agent-executed, stablecoin-settled transactions addresses one of the largest unsolved problems in the new commerce stack.

What I’m Watching

Highest-conviction opportunities in agentic commerce:

• Protocol infrastructure—neutral cross-protocol orchestration; build with UCP in mind
• AI attribution—independent measurement of AI-influenced commerce is urgent, unsolved, and multi-billion-dollar
• Agent-ready commerce middleware—tools making brands machine-readable and context-rich (the Shopify app ecosystem of the agentic era)
• Settlement infrastructure—MPC wallets, stablecoin rails, and dispute resolution for high-frequency agent-mediated transactions

The Punchline

We are roughly where mobile commerce was in 2010: trajectory obvious, infrastructure nascent, incumbents simultaneously threatened and advantaged. The difference is speed: mobile took a decade to reach 40% of e-commerce. Agentic commerce, with McKinsey projecting $5 trillion in global volume by 2030 and Morgan Stanley estimating $190–385 billion in US e-commerce through agents by the same year, will compress that timeline dramatically. The brands, platforms, and infrastructure companies that build for agents today—not for humans using AI as a tool, but for AI as the buyer—will define the next era of commerce. Everyone else risks building beautiful storefronts that no one, and no agent, ever visits.