The agentic web is here

AI agents are beginning to reshape web access, commerce, and publishing, with OpenAI’s GPT-6 Astra reportedly reaching a 73% success rate on the OSWorld 2.0 computer-use benchmark. The shift could improve convenience and accessibility but may also concentrate control in major platforms, threaten ad- and subscription-funded businesses, and create security risks; one cited study found users were about 40% less likely to click through to websites when Google AI Overviews appeared. The article describes an early, uncertain transition rather than a specific near-term market event.

Analysis

The economic fault line is control of discovery and checkout, not agent downloads. If users delegate search and routine transactions, platforms that monetize clicks, sponsored placement, or marketplace tolls face disintermediation; agent-facing ad inventory is not yet a proven replacement, particularly if agents discount sponsored results. GOOG is exposed if agents reduce high-intent search activity, though its own agent and data assets are meaningful offsets. META may have a relative advantage if it can make its assistant a consumer gateway and keep recommendations inside its properties—but that advantage depends on repeat usage and trusted commerce conversion, not launch attention. AMZN’s ability to restrict third-party agents and steer users to its own shopping agent may defend monetization near term, while risking customer friction if alternatives become materially better. EXPE, DASH, CART and ETSY are two-sided: connectors can preserve transactions, but agents may intensify price comparison and weaken customer ownership and take rates.

Near term, this is a narrative rather than an earnings signal: the article provides no traffic, conversion, or revenue attribution. Over 1–3 months, watch agent referral volumes, platform access policies, and evidence of completed transactions. Over 6–18 months, the key question is whether open connectors support direct commerce or whether leading assistants become new toll-taking gatekeepers. Security failures, user distrust, or antitrust intervention could slow adoption or advantage tightly controlled ecosystems. The contrarian point: easier discovery does not automatically mean lower platform power; control may simply migrate from search and marketplaces to the agent layer.

Key Decisions for Investors

• Consider a small, defined-risk 3–6 month relative-value position: long META versus short GOOG, sized modestly rather than as a high-conviction earnings trade. Thesis: assistant-led discovery could pressure search referrals while rewarding a scaled consumer gateway. Reassess if GOOG reports resilient search monetization/referral traffic or META cannot show sustained assistant use and commerce conversion.
• Do not short AMZN solely because it blocks outside agents. Treat its own agent and access controls as a near-term defense; revisit only if evidence shows customers are shifting transactions away from Amazon or restrictions are degrading conversion. Track third-party agent access and Amazon’s own shopping-agent engagement.
• Keep EXPE, DASH, CART and ETSY on a watchlist rather than taking a basket position: agent integration may retain bookings/orders, but could shift bargaining power and compress economics. Require company-level disclosure or credible third-party data on agent-sourced traffic, conversion, repeat rates, and take-rate/margin impact before underwriting winners.
• Use security incidents, permission/checkout failures, or new platform restrictions as adoption catalysts in either direction. Falsifiers for the broader disintermediation thesis include stable search referral economics, agents remaining primarily an information layer rather than completing purchases, or users rejecting delegated transactions; verify these against reported metrics rather than download headlines.