A Benchmark of What Is Real, Not What Is Promised
Most agentic-commerce coverage runs on press releases and roadmaps. AstraWorks took a different tack with its Agentic Commerce Capability Index, which it bills as the first benchmark of observable agentic commerce capability across 24 US grocery retailers. The key word is observable. The scores reflect what shoppers and AI agents could actually use during a May 2026 test window, not internal roadmap activity or aspirational launch dates. That methodological choice is the whole point, because in this market the gap between what is announced and what works is vast.
We welcome this kind of accountability. The retail-AI conversation has been drowning in vendor optimism, and a buyer-side index that grades live capability is a useful corrective. It reframes the question from whether a retailer talks about agentic commerce to whether an agent can find its products, build a cart and complete a purchase today. For CIOs being asked to fund the next round of commerce-AI spending, a benchmark that names the gap is worth more than another trend report.
Amazon Alone at the Top
The scoreboard is lopsided. Amazon scored 63 out of 100, the only retailer to clear fifty. Walmart came second at 49, a 14-point gap behind the leader. Then the field compressed hard: Albertsons and Wegmans tied at 38, Kroger sat at 33, and Target landed at 26. After that, 14 retailers clustered between 10 and 24, and four scored below 10, two of them at zero. This is not a tight race with a leader, it is a leader and a crowd.
The distribution tells the strategic story better than any single score. A handful of operators have built real agentic capability, a slightly larger group has started, and the majority are effectively invisible to AI agents. For the laggards, the risk is not that they lose a feature war. It is that as more shopping intent routes through agents, products that an agent cannot see, parse or buy simply fall out of consideration. Discovery is quietly becoming a pass-fail test, and most of the field is failing it.
The Instacart Dependency Problem
The most consequential finding is structural. Seventeen of the 24 retailers had no customer-facing AI shopping assistant of their own. Of the 18 retailers reachable through ChatGPT, 16 connected via Instacart integrations, with only Walmart and Target holding direct connections. In other words, most major grocers are present in the agentic channel only because an intermediary carries them there. The report flags eleven retailers as platform-dependent, a polite term for renting your own visibility.
We think this dependency is the quiet crisis inside the index. Relying on Instacart to be seen by AI agents is convenient and fast, but it cedes the customer relationship, the data and the margin to a third party at the exact moment those things become most valuable. A grocer that lets an intermediary own its agentic presence is repeating the mistake many made with marketplaces a decade ago, trading short-term reach for long-term leverage. The retailers building first-party capability are paying more now to avoid being a commodity later.
Scale Is Not the Gate
One line from the report deserves to be pinned above every retail-strategy meeting: "Wegmans (~$12B in revenue) scored the same as Albertsons (~$80B). Scale is not the gate. Choices are." A regional grocer with a fraction of the revenue matched a national giant on observable capability. That demolishes the comfortable excuse that agentic readiness is a problem only the largest balance sheets can solve. It is a problem of focus and decision-making, not of budget alone.
There is also a monetization dimension that should worry the field. Amazon was effectively alone in monetizing its assistant, surfacing shopping ads inside the AI experience, while 23 other retailers scored zero on that measure. So Amazon is not just better at agentic commerce, it is already turning the channel into a revenue line while rivals are still trying to show up in it. That combination, capability plus monetization, is how a lead compounds. We would treat this index less as a snapshot and more as an early warning.
What Executives Should Do With This
The actionable takeaway is to test your own observable capability the way AstraWorks did, rather than trusting an internal roadmap deck. Can an agent discover your products, understand their attributes, build a cart and complete checkout right now? If the honest answer is that it works only through Instacart, you have a dependency to manage, not a capability to celebrate. The index gives executives a vocabulary for that conversation and a benchmark against named peers.
We would resist two overreactions. The first is panic-buying a customer-facing assistant just to move up a leaderboard, because a bad first-party agent is worse than a competent integration. The second is complacency, assuming the channel is too small to matter yet. The right posture sits between them: invest in clean, structured, real-time product data and direct agent connections now, because those are the foundations every higher score in this index was built on. The retailers that score well share less a technology than a decision to take the channel seriously before it was obviously urgent.
