互联网与电子商务：代理式人工智能与更具效能的互联网前景

Agentic AI andthe promise of a more Industry Overview TheAgentic Era moving from concept to reality 28 May 2026 The emergence of ChatGPT, Gemini, and Google AI Mode is changing how consumersinteract with the Internet, with conversational AI increasingly becoming a primaryinterface for discovery and decision-making. Agentic AI will build on this evolution,advancing Internet gateway (search) capabilities from information retrieval to evaluationand automated task execution. We think of agents as an embedded productivitysoftware tool for consumers, with the potential to significantly enhance decision quality EquityUnited States Justin PostResearch AnalystBofAS+1 415 676 3547justin.post@bofa.com Nitin Bansal, CFAResearch AnalystBofAS+1 415 676 3551nbansal7@bofa.com Marketplaces will stilloffer significant consumer value AI cannot securedifferentiated supply, deliver products in one hour, or pass on networkeffect cost savings. As users become more accustomed to AI interactions (Google AI Modenow has 1bn users), Online travel and eCommerce marketplaces will need to focus on twoareas: 1) Competitive front-end AI experiences to maintain direct traffic, and 2) Maximizingproduct differentiation, especially in the physical world. We think the best selection, prices,shipping speeds and customer service should be equally attractive to agents as they are 3P: Third PartyAI: Artificial IntelligenceGen-AI: Generative AIMAU: Monthly Active Users New ad formats & toolsto reshape digital advertising Recently, Google I/O and Marketing Live provided an early look at several AI-relateddigital advertising products that either leverage AI use cases, or are designed to enableAgentic transaction capabilities. Three notable initiatives included: 1) Conversational AIads that integrate directly within AI-generated responses, allowing advertisers to engageusers in context with tailored messaging, 2) AI powered shopping ads that surfacerelevant products alongside AI-generated explanations to guide evaluation and decision- Googlein the pole position for the agentic era Wethink Google is well positioned to lead in agentic adoption given leading foundationmodels, agentic time to market advantages, and deep merchant data. Google’s earlymover advantage could help establish ecosystem lock-in and guide behavioral defaultsaround agentic workflows. Moreover, Google’s advertiser ecosystem could see upsidefrom higher AI driven volumes, ad click rates and ad targeting (see page 17). MaintainBuy. OpenAI, Meta and Amazon also have large consumer audiences and deep AI BofA Securities does and seeks to do business with issuers covered inits researchreports. As a result, investors should be aware that the firm may have a conflict ofinterest that could affect the objectivity of this report. Investors should consider thisreport as only a single factor in making their investment decision. AI reshaping consumer behavior The emergence of Generative AI sites and evolution of AI Search is driving a structuralshift in consumer behavior, with AI-native interfaces emerging as the primary gatewayfor consumer traffic and reshaping how users discover, evaluate, and transact online.ChatGPT has scaled to ~900mn weekly active users and processes billions of queries We believe the growing prevalence of AI-driven interactions has the potential to drive afundamental shift at the consumer level, with users increasingly relying on AI surfaces tosynthesize information, compare alternatives, and guide decisions in real time. Theentire Internet ecosystem, from information to transaction sites, will need to evolve user Agentic AI & the next era of Internet utility What is Agentic AI? Agentic AI represents the next evolution of the Internet, capable of independentlyplanning, decision-making, and executing multi-step workflows to complete user-definedgoals. Unlike first-generation AI assistants that primarily generate answers, summarizeinformation, or surface links in response to prompts, agentic systems focus on taskcompletion. Effect Agentic AI assistants will be able to break high-level objectives into Potential consumer agentic adoption timelinesThe Internet is well into the adoption cycle of first-generation AI assistants, while simultaneously entering the experimentation stage for agentic systems. Over the pastyear, consumer AI usage growth has been driven by chat-based AI assistants thatsummarize information, answer questions, assist with coding, and augment search. Theevolution toward fully agentic systems will be a multi-year process (we are close to year •First-generation AI assistants (Started in 2023):During this phase, AIassistants became widely adopted as productivity and knowledge tools. Mostusage centers on chat-based interactions. Personalization and real-world •Early agentic experimentation (started in late 2025):Assistants beginincorporating tool use and limited execution capabilities, but supervision remains essential. Systems can access