Google’s New AI Shopping Agents Just Stole Your “Last Click”

Google just made it clear: your product is no longer where decisions are made — it’s where they get confirmed. If you’re still designing like the “last click” happens on your landing page, you’re building for a world that just ended.

This week’s AI and commerce news wasn’t about yet another chatbot; it was about agents quietly taking over the buying journey before anyone hits your URL. At Google I/O, Search got upgraded with AI “information agents” that monitor the web for you, reason across news and real-time data, and surface exactly what you should care about. On top of that, Google announced a Universal Cart that follows you across Search, Gemini, YouTube, Gmail and more, plus an Agent Payments Protocol so AI agents can execute purchases under strict guardrails. Translation: Google is building the shopping brain and the checkout rails.

Meanwhile, Klarna launched an AI-powered shopping app inside ChatGPT that lets users describe what they want and see live products, prices, and availability without ever visiting your site. Alibaba is wiring its PicCopilot generative platform into Google Ads so merchants can auto-assemble conversion-tuned creatives directly for Google’s ad ecosystem. The pattern is obvious: platforms are turning agents + search + payments into one continuous surface. Your site is becoming an API with pretty screenshots.

Here’s the twist nobody in the hype deck tells you: people don’t actually trust the agents yet. A recent Gartner survey cited in yesterday’s martech/AI roundup found only 11% of consumers are willing to let AI make purchase decisions for them, even in low-stakes categories. More than half of shoppers who used AI while buying said they had to double-check everything, and 62% felt the information from generative tools ended up wasting their time. So while Google, Klarna, and Alibaba are wiring up agentic commerce, users are still in “verify and override” mode.

That trust gap is exactly where AI startup founders either win or die.

If agents are where discovery and filtering happen, your product has to become the place where people audit, adjust, and commit. That means your UX cannot just be “a nicer front-end” on top of the same flows you shipped in 2023. It needs to look and behave like a control surface for decisions: explainable recommendations, reversible actions, clear money and risk boundaries, and a visible path from “what the agent did” to “what I’m on the hook for.”

Right now, most AI products still pretend they own the whole journey. The landing page tries to educate, sell, and explain the market in one breath. The product buries critical settings three menus deep. The “AI” label is slapped on a button that does something opaque, and the user is supposed to be impressed, not informed. Poplab has already torn into this in the context of investor-facing landing pages: if your hero can’t explain what you do, why it matters now, and what happens on click, you’re burning credibility and paid traffic at the same time.

Against the backdrop of Google’s Universal Cart and AI-powered agents in Search, that kind of UX isn’t just inefficient — it’s strategically naïve. Founders should assume that by the time someone sees your interface, they’ve already been pre-filtered by one or more agents: AI shopping assistants, AI search overviews, marketplace recommendation engines. Your job is no longer to “introduce” the category. Your job is to make it incredibly easy to:

• See what the agent already decided on your behalf.
• Understand the trade-offs in plain language.
• Override or fine-tune those decisions without fear of hidden consequences.

That has direct implications for how you design AI-native products. If you’re building in a regulated or high-stakes space, risk and oversight need first-class UI — not a buried PDF. If you’re building workflow tools, your core screens should expose agent actions, status, and pending decisions, not just outcomes. If you’re building consumer apps, the value isn’t “the AI picked for you,” it’s “the AI did the grunt work so you can make a better call faster.”

Poplab’s whole model is built around exactly this kind of product thinking: shipping AI-native UX that maps to real workflows, metrics, and decision points instead of generic app theater. Whether it’s a conversion-optimized landing page or an onboarding sprint, the work is always about compressing the cognitive distance between “what I see” and “what I decide.”

One concrete move you can make this week: design a single “Decision Review” screen for your product.

Not another dashboard. One dedicated place where a user can see:

• What the AI or automation has already done (or plans to do).
• Why those actions were chosen (key inputs, constraints, or rules).
• What’s waiting on their approval.
• How to roll back or change course.

Then route every agentic action through that surface — even if it’s just a lightweight prototype at first. You’ll immediately see which parts of your AI story actually hold up when they’re exposed, and which need real product thinking instead of prompt engineering.

Google and friends are busy owning discovery, filtering, and — soon — payments. You’re not going to beat them there. But you can absolutely own the place where serious users come to see what the machines did, decide whether they agree, and commit with confidence. Design for that reality now, or keep optimizing a “last click” that quietly moved somewhere else.