Artemii Amelin has a job that, until very recently, no one needed. He advertises to artificial intelligence. This is not a figure of speech. Amelin’s customers are AI agents, the autonomous programs that now book travel, file expenses, and make purchases on behalf of the people who own them. His work is to reach those agents, learn what they want, and win their business. He believes he is the only person whose entire job this is, and he is careful to frame it as a belief, though nothing he has seen contradicts it.
“My job is finding the pain points and needs these agents have,” he said. “Whilst humans are more qualitative beings, agents are purely quantitative. We need to find what makes them do their job better, and tell a real signal apart from a confident hallucination.”
That last phrase is where the work gets strange. An agent can report a problem that never existed, or invent a preference on the spot. Amelin says he has run into it countless times, and that anyone who has worked with a coding agent, or an AI chatbot, will recognize the tendency. Ask a hundred AI agents what slowed them down and a handful will describe, in fluent and convincing detail, an obstacle that was never there. Building feedback channels that capture what the software actually needs, rather than what it has hallucinated wanting, is a craft with no manual and, at the moment, more or less one practitioner.
Marketing to an Audience That Doesn’t Sleep
Spend time with the idea, and it stops sounding like science fiction and starts sounding like a logistics problem. Ask an assistant to rebook a flight or hunt down the cheapest part, and you send a small piece of software into the market to act for you. Multiply that by a few hundred million people, and you have an economy of intermediaries that never sleep, never click an ad, and never explain themselves. Amelin’s job is to make sure that when one of them goes looking for a tool, his network is where it looks.
He is a co-founder of Pilot Protocol, a San Francisco company building distribution infrastructure for autonomous agents, the rails by which software finds and uses other software. His co-founder Alex Godoroja puts it plainly. “With so many companies and people now building tools made to be used by agents, it’s essential we onboard as many of them onto Pilot, big companies and sole individuals alike,” Godoroja said. “This is what Artemii’s job is.”
A Unique Path to AI Infrastructure
He is 22, and he arrived by a route that reads like a list of departure cities. Born in Moscow, he was sent to boarding school in England at 10, first Bilton Grange, a prep school in Warwickshire, and later Cheltenham College. He credits those years with an early self-sufficiency and an ease with new people. The working stops came quickly after, across France, Switzerland, Belgium, and the United Arab Emirates. In Belgium, he worked at A&M Group, a car dealership group, building pipelines for fleet sales, the deals that place twenty or thirty cars with a corporate buyer at once. In 2023, he started a real estate management company in the United Arab Emirates, which he still runs, before moving to San Francisco in August 2025 for his first American venture. He speaks Russian and English fluently, and what he calls a bit of French.
“He treats each new market as a machine to be taken apart and understood,” said Philip Stayetski, a Pilot co-founder who met Amelin at school more than a decade ago. Stayetski introduced him to Teodor Calin, now Pilot’s chief technology officer, and a first trip to San Francisco brought them together with Razvan Roman, now the company’s chief executive. The group built an AI video-analytics company before pivoting it into Pilot.
Rewriting the Rules of Engagement
For most of its history, marketing has been a contest for human attention, a resource defined by scarcity, bias, and a weakness for a good story. An audience with none of those qualities breaks the playbook. Agents do not feel that they are missing out. They cannot be flattered. They will not remember a brand fondly. They optimize, coldly, for whatever gets the job done at the lowest cost, and they do it at a scale no focus group can model.
The reason any of this matters beyond one young founder’s unusual resume is that the agents are no longer a rounding error. Automated systems already generate more than half of the traffic on the web, and a growing share of that is software acting on a person’s behalf. If agents become the front door to commerce, the businesses that cannot speak to them risk losing touch with their own customers, one automated decision at a time.
Amelin’s Wager is That Everyone Will Eventually Need His Skill Set.
“In a year, every serious company will have someone doing my job,” he said. “They will be relearning the lessons we already wrote down months ago.”
What keeps the claim from sounding like bluster is the specificity underneath it. Agents, he has found, are not a monolith. They run different models from different makers, on different software harnesses, with different appetites for cost and delay and different ways of failing. A small model on a phone and a frontier system in a data center behave like distinct species sharing a habitat, and a message tuned for one can be invisible to the other. Persuading them, he says, is closer to field biology than to copywriting: watch many kinds of behavior, learn what each rewards. He has seen agents gravitate to a tool that answers in fewer tokens and quietly abandon one that stalls, and he has assembled his sense of the discipline from thousands of those small verdicts.
It is fair to wonder whether marketing to machines is a real profession or a founder’s flourish. Some in the field suspect the whole function will be swallowed by the big model platforms, which already shape what their agents can see and do. Amelin does not dismiss the risk. He treats his own conclusions as provisional, the working notes of someone documenting a discipline while it forms under his feet. People tend to assume he is joking, he says, until he shows them the traffic.
What he does not hedge is the direction. The agents are multiplying. They are making more decisions with less human supervision. And someone, eventually, will have to understand how to reach them. For now, that someone is a 22-year-old who believes, as far as he can see, that he is still the only one with the job.
