Amazon staffs AgentCore team to plug external AI agents into the marketplace

What's happening

On May 11, 2026, Amazon was reported to be staffing a new team focused on connecting its marketplace and ads stack to external AI agents (ChatGPT, Perplexity, Claude, Gemini). The hook is a public Amazon job posting for a Principal Product Manager - Technical assigned to "External Services, AgentCore", first surfaced by Dataconomy. We pay attention to moves in this category because they reshape the numbers brand managers see in the cockpit each morning. We pay attention to moves in this category because they reshape the numbers brand managers see in the cockpit each morning.

The role description is explicit. The hire will own APIs and protocol surfaces that let third-party agentic platforms transact against Amazon services (Amazon.jobs listing #3185564). Pair that with the May 7 release of Amazon Bedrock AgentCore payments and the picture is clear: Amazon is building the rails for outside AI agents to discover, price, and buy on Amazon - and to settle payment without a human session.

For sellers and agencies, this is bigger than another Rufus update. Rufus is Amazon's own agent. AgentCore External Services is the API that lets everyone else's Agents enter the funnel.

Timeline of the past two weeks

Why this matters for Amazon sellers

Until now the agentic-commerce conversation on Amazon has been Rufus-shaped. Rufus reads listings, summarises them, and increasingly recommends or pre-fills carts inside Amazon's own surfaces. AgentCore External Services flips the direction: a buyer talking to ChatGPT or Perplexity could trigger an Amazon purchase without ever touching the Amazon app or website.

This is the same pattern that played out with Amazon vs Perplexity Comet and Gemini's agentic checkout on Shopify and Walmart - except this time Amazon is the one opening the door rather than litigating it shut.

Who else is building this rail

What you should do this month

1. 1

   Treat structured listing data as the new SEO

   Agents do not scroll. They read attributes, A+ structured fields, BSR, review velocity, and price history. Audit every flagship ASIN for completeness using Products Feed.

2. 2

   Re-baseline conversion attribution

   Tag a "direct / unattributed" segment in your reporting now so you can isolate agent-driven orders as the volume ramps. Build it once in Custom Breakdowns and refresh hourly.

3. 3

   Pressure-test PPC ROAS assumptions

   If a meaningful share of demand routes through outside agents, your branded PPC campaigns may be paying for clicks the agent would have earned organically. Watch PPC Analytics for branded-term ACoS drift over the next two quarters.

4. 4

   Plan a margin floor that survives an agent funnel

   Agents optimise for buyer outcome, not seller margin. Reconcile every 40+ Amazon fee type at the SKU level in Profit & Loss so you know the actual floor before an agent-driven price war finds you.

How Nova helps

Nova is the operating system for Amazon brands. Three capabilities are directly relevant as agent-driven demand ramps:

• Amazon Data API - delivers analysis-ready Amazon data to BigQuery, Snowflake or Databricks so your team can join order data with whichever agent-platform telemetry surfaces over the next 12 months.
• Amazon Data Library - 200+ Amazon metrics and 40+ fee types so margin and conversion math stay honest as new buyer surfaces appear.
• Seller Cockpit - daily ASIN-level Buy Box, returns, and conversion view so an agent-driven shift in mix is visible the day it happens.
• Agencies - if you manage multiple brands, the unified P&L and saved views handle the cross-account reporting that agent-era reporting will demand.