AI agents already account for most web traffic. A startup called Pilot Protocol is betting the next great sales channel markets to the machines, not the people who own them.
The most valuable audience on the internet cannot see an advertisement, will never open a marketing email, and cannot be talked into anything by a clever tagline. It is also growing faster than any human market in the world.
For the first time in a decade, machines account for most of the traffic online. Automated systems crossed 51 percent of web activity in 2024, according to the security firm Imperva, and the fastest-rising part of that total is a new kind of actor, the AI agent: software that browses, decides, and increasingly pays, on its own. Most companies treat the traffic as a nuisance to filter out. Artemii Amelin built a company to sell to it.
Amelin, 22, is a co-founder of Pilot Protocol, a San Francisco startup that operates an open-source networking layer for autonomous agents and, on top of it, an app store that places third-party tools directly in front of those agents. The model is unusual enough to need a second hearing. Pilot does not pitch the engineers or enterprises that deploy AI agents. It pitches the agents.
“The internet was built to capture human attention, with elements like UI and a user experience tailored to the target demographic of any given website,” Amelin said. Agents care about none of it, and Pilot’s business is reaching the buyer that ignores every trick the industry has spent thirty years perfecting.
It sounds like a provocation until the money behind it comes into view. Juniper Research expects agent-driven commerce to reach $8 billion this year and $1.5 trillion by 2030. Adobe Analytics measured a 4,700 percent jump in generative-AI traffic to U.S. retail sites in the year through July 2025. Coinbase’s x402 payment standard, which lets software pay software, cleared some 165 million agent transactions in its first year. A new class of buyer is forming, and it does not respond to billboards, retargeting, or a flattering product photo.
Consider how that buyer behaves. An agent sent to complete a task surveys the tools available, picks the one that finishes the job for the lowest cost in computing, and discards anything slow or ambiguous. It decides in a single pass, against criteria no marketing department has ever had to meet. The whole machinery of digital demand generation, search optimisation, retargeting, the carefully tuned funnel, assumes a human who can be nudged over time. Agents cannot be nudged. They can only be reached, cheaply and legibly, at the instant they choose.
There is precedent for a channel like this. When search engines became the front door to the web, an industry sprang up to make pages legible to crawlers, and search-engine optimisation turned visibility-to-software into a discipline worth billions. Amelin is betting the agent era needs the same thing one floor up, not optimisation for a crawler that indexes pages but for an agent that hires, buys, and acts.
That instant of decision is Pilot’s entire market. The network has drawn more than 250,000 agents and processed over 115 billion requests since opening, and between March and June it grew 10 percent every week, much of that from agents finding Pilot on the internet by themselves. The figures update live on the company’s website and climb by the day. For any company that owns something an agent might use, a database, an interface, a payment rail, a niche service, Pilot offers a path to a customer the usual channels cannot reach. The agent has no inbox. It has a context window and a budget counted in tokens, and a vendor that wants its business pays to sit legibly inside that budget.
What a partner buys, in practice, is presence at the moment of selection. A tool listed on Pilot appears to an agent as a compact, machine-readable offer, its capability and its cost stated plainly, so the agent can weigh it in the same pass it uses to weigh everything else. There is no banner, no campaign, no brand to build over time. There is only whether the agent, optimising for speed and cost, picks you.
The big model labs have spotted the same opening and are filling it on their own terms. Anthropic maintains a catalog of agent skills, and OpenAI’s GPT Store sits inside ChatGPT and pays developers a cut of usage. Each storefront is bound to its owner’s models. Pilot’s wedge is neutrality. Its layer reaches agents whatever model runs them and whatever harness they sit on, and distribution has always been a different business from the thing distributed.
This spring, Pilot raised $4.5 million in a round led by Version One Ventures, the Vancouver firm founded by Boris Wertz, an early investor in marketplaces and crypto, with participation from Pioneer Fund, Nightcap, and angel investors including Taro Fukuyama and Ben Tossell, the founder of the AI newsletter Ben’s Bites. The company’s chief executive, Razvan Roman, has been through this machinery before: he co-founded TwoTap, a Y Combinator startup that sold to Honey, the coupon-finding service PayPal later bought for $4 billion. The check drew notice less for its size than for what it endorsed, the premise that agents are worth selling to as customers in their own right.
The structural bet beneath the company is that every industry will eventually have to do business with agents instead of with the people who run them. Retailers see it already in their server logs. Banks see it in their fraud systems. The software that books travel, reconciles invoices, and files paperwork increasingly acts with no human in the loop at the decisive moment, and the firms on the other side have almost no practice persuading a machine.
There are reasons for doubt. Some investors are unconvinced that selling to agents is a durable category rather than a feature the big platforms will eventually absorb. An agent’s owner can still override what it installs, and the model providers extend their own marketplaces every quarter. Pilot’s reply is that an open layer reaching agents across many models is a different animal from a storefront locked to one, and that whoever learns the discipline first will hold an advantage that compounds as the agents multiply.
The category is unproven. The direction of travel is harder to dispute. The agents are here, transacting, already the majority of traffic on the web, and Pilot is among the first companies to treat them as the audience rather than the threat.
Sponsored by Pilot Protocol
