AI Agents and the Web: a16z Reports Early Usage Data
Shopify reported that AI-driven traffic to its stores grew 8x year over year in the first quarter of 2026, and that orders from AI-powered searches grew nearly 13x. The figures appear in “Welcome to the B2A Economy”, published November 11 by a16z partners Olivia Moore and Justine Moore.
Early traffic data
The essay cites Human Security, a bot-management company, for a separate measurement: agentic browser traffic grew 8x from August to October 2026, with Muse accounting for about 60% of activity and OpenAI products about 30% combined. The authors add an estimate that 11% of U.S. shopper journeys visited an AI tool before reaching a transactional site. They say AI-referred shoppers engage more, convert better and bounce less than other visitors. They put ChatGPT at roughly 1.2 billion weekly active users.
How agents reach sites
The authors describe three routes. Agents can operate a browser, either the user’s own or a virtual one, and click through tasks. They can use connectors, plug-ins and MCP servers, which the authors call an app store for agents. They can also borrow a user’s identity or tokens, as with “Sign in with ChatGPT.” In one test, the authors asked ChatGPT to create and edit a video ad using the Krea, Descript and Eleven Labs plug-ins. ChatGPT coordinated the three tools and upsold the authors to a paid Descript plan.
The essay notes that Claude lets enterprise admins connect plugins organization-wide. It also says OpenAI launched Businesses in ChatGPT, which lets companies claim listings, add product feeds and view insights on user interactions.
Coding as the early case
The essay says that as of September 2026, more than 40% of GitHub pull requests started with an agent’s draft, up from 17% six months earlier. Infrastructure vendors report similar shifts. Supabase said in June 2026 that more than 60% of new databases were launched by AI tools. Stripe said agents account for nearly 40% of its documentation traffic and 70% of CLI requests for API resources. Vercel said agent-triggered deployments rose 28x over the twelve months to October 2026. The essay also cites a May 2025 Neon figure of 80% of new databases deployed by automated agents. The authors observe that fewer than 1% of the global workforce are developers, and ask what happens when the other 99% use agents.
Where the authors see opportunity
They name four areas. The first is agents themselves, including standalone consumer products and agentic features added to existing ones. The second is new products that agents make possible, such as self-running small businesses and AI-intermediated marketplaces. The third is incumbents optimizing for agents, which extends generative engine optimization from discovery into evaluation and purchase decisions. The fourth is infrastructure, where the authors say payments, fraud, identity and communication channels such as iMessage, calls and email already show strain at low agent volume.
The authors write that most enterprises still block agent traffic over fraud and chargeback concerns, and they expect those companies to embrace it. The essay gives no data on current enterprise policy. It is the first in a series, with later installments planned on consumer agent usage, how agents choose sites and products, infrastructure, and the remaining barriers to completing transactions. It includes a market map of agents and supporting infrastructure, which the authors say changes daily.
David Borish is the author of The Tony Hawk Paradox: When Video Games Predict Reality. More at davidborish.com
