Andreessen Horowitz partners Olivia Moore and Justine Moore argue in a November 10 essay, “Welcome to the B2A Economy,” that businesses should start treating AI agents as customers, not only as threats to block. The essay’s most quoted line says agents “now make up the majority of web traffic.” The harder number underneath is narrower, and better documented: HUMAN Security, the essay says, “saw agentic browser traffic grow 8x between August and October,” with Meta’s Muse taking the largest share (about 60%) and OpenAI products a combined 30%. That 8x figure measures agent requests on HUMAN customer sites, not all traffic on the public web.
Quick Take
- The argument: As agents move from answering questions to completing tasks, they will choose products, subscriptions and where to spend money. Firms that spent years blocking bots may have to compete for that traffic.
- The 8x number: a16z cites HUMAN Security for an eightfold rise in agentic browser traffic from August to October 2026, with Muse around 60% and OpenAI products around 30% of that agent traffic.
- The majority claim: The sentence that agents already are “the majority of web traffic” is not given a source in the essay. A nearby chart in the piece uses broader “bot versus human” figures, which include crawlers and other automation, and are not the same thing as AI agents.
- Other cited figures: Shopify, per the essay, saw AI-driven store traffic grow eightfold year over year in the first quarter of 2026, and orders from AI-powered searches nearly thirteenfold. The authors also write that, as of September 2026, upwards of 40% of GitHub pull requests started with an agent’s draft, up from 17% six months earlier.
- Who is affected: Retailers, travel and media sites that already see agent traffic, and any company still defaulting to “block all bots.”
In plain English
A well-known venture firm says businesses should start selling to software agents, not only to people, because those agents are starting to shop, book and click on our behalf. One data company it cites saw traffic from AI agents in browsers grow eight times in two months, mostly from Meta and OpenAI tools. That is not the same as saying most of the entire internet is now AI. A lot of older “bot” traffic is just search crawlers and scrapers. The practical question for a company is more ordinary: when an agent visits your site, do you shut the door, or try to be easy to buy from?
What the essay says
“The primary users of the Internet are no longer human. They’re AI agents,” the Moores write. They describe an early platform shift: AI has already changed what people can do, they argue, but “its transformation of distribution is just beginning.” If agents start completing tasks, “they’ll increasingly decide which products to use, which services to subscribe to, and where to spend money on our behalf.”
The supporting examples mix first-party company reports and third-party telemetry. On the consumer side, the essay cites Shopify’s first-quarter 2026 figures (eightfold AI-driven traffic growth year over year, nearly thirteenfold growth in orders from AI-powered searches) and says AI-referred consumer traffic converts better and bounces less than other traffic. It also cites an estimate that, even before agents, about 11% of U.S. shopper journeys visited AI before a transactional site.
On the builder side, the authors point to coding agents as a preview. They write that upwards of 40% of GitHub pull requests started with an agent’s draft as of September 2026, versus 17% six months earlier. They also relay vendor figures: Supabase, in June 2026, saying more than 60% of new databases are launched by AI tools; Neon, in May 2025, saying automated agents deploy 80% of its new databases; Stripe, in June 2026, saying agents accounted for nearly 40% of documentation traffic and 70% of CLI requests for API resources; and Vercel, in October 2026, saying agent-triggered deployments had increased 28 times over twelve months. Those are the vendors’ own numbers, as presented by a16z, which also notes that some information in the essay comes from third parties including portfolio companies.
The authors’ market map includes standalone consumer agents, incumbents that optimize for agents, new products that could exist only with agents, and infrastructure for payments, identity and communications. They mention that OpenAI had just launched Businesses in ChatGPT, which i10x has already covered in ChatGPT Business Hub.
What the HUMAN numbers do and do not show
HUMAN’s agentic-traffic figures, as used by a16z, are the essay’s best empirical hook, and they need a tight frame. They describe observed agent requests on HUMAN’s customer sites, including allowed and blocked activity, not a census of the whole web. Meta’s Muse and OpenAI tools dominating that slice is a story about which agents are busy on those sites in the autumn of 2026. It is not a story about all human page views disappearing.
The “majority of web traffic” line is weaker. The essay states it without a citation. Coverage of the accompanying “bot versus human” chart has pointed out that those series count bots in general (crawlers, scrapers, monitoring, abuse), that a 2026 majority figure in such charts is sometimes a forecast, and that agentic traffic can grow very fast from a small base without being most of the web. i10x treats the 8x HUMAN rise as a real, bounded measurement, and the majority claim as an argument ahead of the evidence the essay itself supplies.
United Airlines’ recent decision to block unauthorized agents from booking flights is a reminder that many firms still see unattended agents as a liability, not a customer. The a16z essay says that tension is the point. For years, the authors write, retailers and platforms built walls to keep bots out. Now, they argue, those firms will have to fight for the bots’ business.
Pros and cons
For consumer businesses
- Pro: If agents already refer higher-intent traffic, making product data and checkout easier for them to use could be a new distribution channel.
- Con: Opening the door without fraud, identity and refund rules built for agents could raise chargebacks and account takeovers.
For infrastructure and software firms
- Pro: Coding-agent traffic is already a large share of some developer platforms’ signups and API use, according to the vendor figures the essay cites.
- Con: Those figures are self-reported and come, in some cases, from companies in a16z’s orbit.
For readers of market essays
- Pro: The piece usefully separates a distribution story (agents as a new customer) from a model-quality story.
- Con: Its opening claim about “the majority of web traffic” is stronger than the evidence presented for it.
What’s still open
- A public, like-for-like series that separates AI agents from older bots across a representative slice of the web.
- How much agent traffic actually reaches checkout and payment, versus product and search pages.
- Whether firms that welcome agents see fraud rise, conversion rise, or both.
- How “sign in with ChatGPT” and similar identity-lending schemes will be regulated.
About the analysis
Researched and written with AI assistance under the editorial responsibility of the i10X News team. The title image is an AI-generated image. Read our AI Use Policy. Spotted an error? Request a correction.
This independent i10x analysis is based on the November 10, 2026 a16z essay “Welcome to the B2A Economy” by Olivia Moore and Justine Moore, and on the HUMAN Security figures as they are presented there. i10x did not receive a separate data cut from HUMAN or Shopify for this article.
i10x Perspective
The 8x rise in agentic browser traffic on HUMAN’s network is a real signal, and businesses should plan for agents that compare, click and sometimes buy. The mistake is to flatten that into “agents already are the internet.” They are not, not on the evidence this essay provides. The companies that will look smart in a year are the ones that can tell a helpful agent from a scraper, serve the first, and still say no when the second tries to book a flight or drain a loyalty account.
