AI agents disrupt “Here Comes the Wolf”? Bernstein: Market fears are out of touch with fundamentals, and these emerging US e-commerce stocks have been “mistakenly killed”

The Zhitong Finance App learned that the “disruptive narrative” of AI agents (AI agents) against internet platforms has once again taken Wall Street by storm. In the past two trading days, the US internet market platform sector was sold off. The median basket of Bernstein-related stocks fell 7%, and the valuation multiples of targets such as DoorDash, Instacart, and Uber are once again facing torture. However, Bernstein clearly stated in the latest research report that the AI agency business is still in its very early stages, changes in consumer behavior are far from occurring, and there is a significant disconnect between market fears and actual fundamentals, which has created buying opportunities for investors.

Agent Product Intensive Release, Agentic Commerce Narrative Heats Up

The trigger for this round of panic was the intensive launch of a series of consumer-grade AI agent products. On September 8, Meta launched Muse, a personal AI agent, which runs on its self-developed model Muse Spark. It can be used through a standalone app, website, or WhatsApp to send emails, book trips, fill out forms, and complete shopping payments on behalf of users. Muse is connected to Stripe's Link smart wallet, can directly complete checkout at more than 1 million merchants, and use disposable virtual cards to protect users' real payment information. Meta's chief AI officer, Alexandr Wang, told the media that the company is looking at commercial-related fields as a potential source of future revenue.

Silicon Valley's up-and-coming Instinct represents another path. This AI personal assistant, created by 23-year-old founder Noah Shinn, allows users to connect to WhatsApp, iMessage, or email accounts with one click to send tasks via SMS or phone — reply to emails, manage calendars, book air tickets, compare insurance, and even cancel subscriptions. Instinct only opened an invitation-based closed beta in February of this year. On August 26, it completed Series B financing of 250 million US dollars. The valuation soared to 2.5 billion US dollars, and the valuation doubled fivefold within three weeks.

At the same time, OpenAI released GPT-6 Astra, which achieved a score of 59.3% on the Aggregate Last Exam benchmark. It can operate desktop software, fill out forms, and run tests, and was called “the beginning of the AGI era” by co-founder Greg Brockman. Furthermore, the continued popularity of OpenClaw, Claude Cowork's Dispatch function, XAI's Grok Bot, and Apple's plan to restructure Siri with AI capabilities together form a complete narrative puzzle for “always-on” consumer agents.

The gap between fear and reality

Bernstein's core judgment is that the vast majority of current discussions are still theoretical deductions. According to the report, although applications such as ChatGPT and Gemini have become widely popular, recommended traffic from AI chatbots to top e-commerce platforms still accounts for less than 1% of total traffic. According to Similarweb data, in August 2026, Wayfair's Genai recommended traffic accounted for 1.4%, Walmart 1.2%, while DoorDash and InstaCart were only 0.4% and 0.3%, respectively.

Consumer trust is another insurmountable obstacle. According to the “Agentic Commerce 2026” report released by Checkout.com, 33% of consumers expect at least 10% of their purchases to be AI-driven within the next year, yet 25% say they will never entrust purchasing decisions to AI, and 27% say they don't trust any agency to operate AI shopping agents. The average single purchase that consumers are willing to let AI agents complete without additional approval is only £177, which is a gap with the merchant's expectation of £200.

In terms of categories, the shopping scenarios where consumers are most willing to hire AI agents are groceries (41%) and household goods (31%), while financial services are only 15%. Notably, 57% of consumers say they are willing to switch brands if AI agents can identify more cost-effective alternatives — which means agency businesses may reshape the brand loyalty landscape.

The back-end fulfillment “moat” of traditional platforms: vertical differentiation

In his report, Bernstein constructed a framework for evaluating market platform moats, covering the six dimensions of supply uniqueness, purchasing decision complexity, transaction complexity, competitive positioning, customer loyalty, and excess yield. According to this, there are significant differences in the extent to which different vertical fields are impacted by AI agents.

Online car-hailing and takeout delivery are considered the most resilient. Uber's supply-side liquidity advantage, DoorDash and Uber Eats's breadth of choices for restaurant takeout, and Instacart's real-time logistics system in the grocery sector all form structural barriers that are difficult to be easily circumvented by AI agents. The user stickiness brought about by the subscription system further strengthens defense capabilities.

Traditional e-commerce faces a more complicated situation and is relatively vulnerable to the impact of AI's convenient price comparison. Bernstein pointed out that the retail market is highly fragmented, creating space for AI-assisted price comparison and shopping. But asset-light platforms like eBay and Etsy need attention because product discovery is where their core value proposition lies. Wayfair's home furnishing SKU is highly visual and emotional, and logistics capabilities constitute a competitive barrier, but the market structure is still a major concern.

What investors should pay attention to

Bernstein clearly maintained “outperforming the market” ratings for DoorDash (DASH.US), Instacart (CART.US), Uber (UBER.US), and Wayfair (W.US), and gave eBay (EBAY.US), Etsy (ESTY.US), Lyft (LYFT.US), and Zillow (ZG.US) ratings “consistent with the market”. The report emphasizes that the next 12 months will be a key observation window — whether consumers actually shift from “learned behavior” to agent-driven shopping models, and whether market platforms can transform AI capabilities into quantifiable revenue growth.

Bernstein analyst Nikhil Devnani's team likened the current situation to a long-standing debate between Uber and autonomous vehicles: market sentiment is already too negative, and the actual impact of fundamental changes will take time to verify. For patient investors, the value depression created by panic selling may be more noteworthy than AI agents themselves.