The Future of E‑Commerce: Conversational → Agentic
What’s materially new
- Amazon moved conversational shopping into the default experience
Rufus—Amazon’s generative AI shopping assistant—is now available to all U.S. customers in the app and on desktop.1 Internal planning documents reported by Business Insider suggest Rufus could drive >$700M in operating profit in 2025 via “downstream impact.”2 Reviews, however, remain mixed; some usability gaps persist (e.g., lists without links, over-indexing on Amazon-branded items).3 - Walmart went all‑in on agents (not just chat)
Walmart launched Sparky, framing it not as a point feature but as a strategic layer: “Sparky is more than a feature — it’s a foundation for what’s next.”4 The company is rolling out a suite of “super agents” for customers, associates, suppliers, and developers—part of a broader push to make half of sales online in five years.5 Consumer research in Retail Rewired 2025 shows 69% of shoppers still start with a search bar vs. 13% using chat/AI/voice assistants—so search isn’t dead, but the slope is clear.6 - Google’s AI Mode turned shopping into an end‑to‑end agentic flow
At I/O 2025, Google introduced an AI Mode that pairs Gemini with the Shopping Graph (now 50B+ product listings; >2B refreshed every hour), adds agentic checkout (track price → “buy for me” → Google Pay completes checkout), and launches virtual try‑on with your own photo—“upload a full‑length photo of yourself” to see how billions of apparel items look on you.7
Separately, Gemini Live added on‑screen visual guidance—you can point your camera and the assistant highlights the right object, a step toward real‑world “show and shop.”8 - Third‑party agents can now complete purchases in chat
Perplexity partnered with PayPal to enable instant purchases with PayPal or Venmo inside Perplexity Pro—a concrete example of payments meeting agentic search.9 - Messaging channels are standardizing AI workflows for commerce
Meta announced centralized campaigns in Ads Manager across WhatsApp/Facebook/Instagram with expanded AI support and calling for WhatsApp Business—smoother, measurable funnels that start and finish in chat.10 - Storefront‑grade agents became buildable off‑the‑shelf
Shopify’s Summer ’25 release shipped three big building blocks:
- Knowledge Base (first‑party, structured content for assistant answers).11
- Shopify Catalog (“one catalog, millions of merchants”) so apps and AI agents can search live product data across Shopify.12
- Storefront MCP (Model Context Protocol) to let agents search products, answer policies, build carts, and start checkout against a specific store.13
- Research to action: ChatGPT is now an agent
OpenAI launched ChatGPT agent, which “thinks and acts” and uses its own virtual computer to complete multi‑step tasks—browsing, form‑fill, creating files—blurring the line between search, shopping, and doing.14 We also have more proof that AI can shoulder real work: Klarna says its AI assistant did the equivalent work of 700 FTEs within a month of launch.15
What still bites (and how to plan around it)
- Answer quality and incentives. Large catalogs + ad economics can bias results. Build guardrails: require linked SKUs, show sources, and log every recommended item vs. clicked/purchased item to audit bias.3
- Retrieval and freshness. Google’s Graph claims 2B+ hourly updates; your catalog may not. Ship a freshness SLA (price/stock/variant latency), and fail closed when data is stale.7
- Policy & trust. EU AI Act requires that users be told at or before first interaction that they’re engaging with AI; synthetic outputs need disclosure. California’s BOT Act requires clear bot disclosure when influencing purchases. Bake on‑surface disclosures and opt‑in consent for agentic tasks (price tracking, checkout, account lookups).1617
A 90‑day plan to make this real
1) Data foundation (weeks 1–3)
- Stand up a product knowledge base with canonical FAQs, fit notes, compat charts, returns, and warranties—prioritized by top support/search intents. (Shopify Knowledge Base if you’re on Shopify.)11
- Publish an AI‑friendly catalog feed with normalized attributes (size/fit/material, compatibility, region/price/stock) and traceable references back to PDPs. (If eligible, ingest via Shopify Catalog.)12
2) Agentic storefront (weeks 2–6)
- Add a conversational “Ask to Buy” experience to PDPs and Collections. On Shopify, wire Storefront MCP tools for search → compare → cart → checkout and customer‑account tools for orders/returns. Start with 3–5 high‑intent journeys (e.g., “build my kit,” “compare X vs Y,” “reorder with substitutions”).13
- Require linked SKUs in every recommendation and show a compact source panel (reviews, spec tables).
3) Checkout + payments (weeks 4–8)
- If you reach users via agentic search (e.g., Perplexity), pilot in‑chat payments with PayPal/Venmo where available; otherwise deep‑link carts with one‑tap flows.9
- Implement price‑tracking → notify → consented ‘buy for me’ patterns on your site/app to mirror Google’s agentic checkout UX.7
4) Trust, safety, and compliance (weeks 1–ongoing)
- Ship inline AI disclosure (and an “explain this suggestion” link); log decisions for audit.
- Gate sensitive actions (returns, address changes, payment) with strong auth + explicit consent.
- Align disclosures with EU AI Act Art. 50 timing/format and CA BOT triggers (commercial influence).1617
5) Measurement (weeks 2–ongoing)
Move beyond CTR to Agentic Performance Metrics:
- Assist rate (sessions with agent actions), Task completion (end‑to‑end journey done), Time‑to‑decision, SKU coverage, Abandon due to low confidence, Appeals/rollbacks, and Attribution lift vs. baseline search.
- Track bias drift (share of house‑brand recs vs. category share), freshness faults (OOS/price mismatch surfaced).
What to watch next
- Camera‑forward shopping. As Gemini Live spreads visual guidance beyond Pixel, expect “show and shop” to move from novelty to norm for complex purchases (DIY parts, cosmetics shades, fit checks).8
- Agent protocols. Shopify’s MCP servers + Catalog are early proof that agent ecosystems will standardize. Expect more retailers to expose agent endpoints for discovery, price/stock, and order orchestration.1213
- Search → agent shift. Walmart’s own survey says search still dominates, but the slope is toward agents. Budget and measurement models should be ready before your customers (or their agents) are.
