Overview

Rachael Rajan authored a Reuters piece published on 15‑06‑2026 that examined the current state of “agentic” shopping – the use of AI agents to perform parts of the purchasing journey on behalf of consumers. The article referenced live stock movements for the three companies discussed: Amazon (NASDAQ: AMZN) fell 1.23%, Walmart (NASDAQ: WMT) rose 0.45% and Google (NASDAQ: GOOG) rose 0.45%.

Test Scope and Methodology

Bernstein conducted a comparative test of five AI‑driven agentic shopping tools: OpenAI’s ChatGPT, Google’s Gemini, Anthropic’s Claude, Amazon’s Alexa, and Walmart’s Sparky. The evaluation focused on three stages of the shopping funnel – product identification, pricing validation, and purchase enablement – to pinpoint where each tool succeeded or broke down.

Key Findings

  • No tool achieved unsupervised purchase completion. Bernstein concluded that “there is currently no agentic shopping tool capable of buying a product in an unsupervised fashion,” and that human engagement is still required at multiple steps.
  • General‑purpose models (ChatGPT, Claude, Gemini) lack retailer‑specific data. These models rely on web‑scraping and third‑party articles, which leads to occasional failures in SKU identification and pricing accuracy. They also cannot access structured catalogue information, real‑time pricing, inventory levels, or delivery options, causing them to falter on fulfillment and payment handling.
  • Retail‑native agents (Alexa, Sparky) provide precise, near‑transactional experiences but are confined to their own assortments. While they can offer highly accurate product recommendations within the retailer’s catalog, they are unable to complete the transaction within the assistant interface.
  • Emerging integrations show progress but still require manual checkout. ChatGPT’s Agentic Commerce Protocol (ACP) and Claude’s integration with Uber Eats (NYSE: UBER) enable users to place grocery or food orders, yet the final payment must be completed by the customer on the retailer’s website. The same limitation applies to Alexa and Sparky, where transactions cannot be finalized directly inside the voice assistant.
  • Strategic gap identified. Bernstein highlighted that “bridging this gap, from either end, will be key as the technology evolves,” emphasizing the need to combine the flexibility of general AI with the data fidelity of retail‑native solutions.

Implications

The analysis underscores that while AI agents are increasingly capable of independent product selection and recommendation, the absence of embedded payment mechanisms and real‑time retailer data prevents a true end‑to‑end agentic e‑commerce experience. Continued partnerships between foundational AI providers and retailers are expected to address these limitations, but full automation remains a future objective.