文字起こし 🔗
I've spent 6 years studying businesses for Trends.vc.
Lately a new kind of customer shows up in the companies I study: the AI agent.
I grade business models by who pays, from a C for consumers to a B for businesses and an A for agents.
Businesses beat consumers on four counts:
- More urgency
- Higher prices
- Bigger budgets
- Fewer cancellations
Paul English co-founded the travel site Kayak. Then he launched a vacation-planning app called Lola. Within its first year his team found 80% of its users were traveling for work.
Business travelers stuck with Lola, so English rebuilt it as a corporate travel tool.
Nick Gray started Museum Hack by giving offbeat museum tours to his friends for free. Then he added weekday team-building tours for companies. They became one of the company's most successful lines.
The company grew to $2.8M a year and Gray sold it in 2019.
Benjamin Houy grew a French-learning app to $10K a month. Growth stalled after most of his audience bought it.
He felt consumers paid little and canceled fast, so he left the app to build a tool businesses use to track how AI chatbots mention them.
If you sell to consumers today, a business version of the same product can move you up a grade.
Since an agent works in the background, selling to agents beats selling to businesses. Agents search the web and build apps while the people who set them up sleep.
The research firm IDC expects companies to go from about 29 million AI agents in 2025 to over 2 billion by 2030. That's about 1.2 million new agents a day, more than 3 times the number of babies born each day.
If you sell to agents, your pool of possible customers grows by about a million a day.
Nikita Shamgunov runs a database company called Neon. The coding site Replit plugged Neon in and created about 10 databases a day.
Then Replit launched an AI agent that builds apps for people.
Neon created 13,000 databases in one day. Most of them went to people who never heard of Neon.
Ivan Burazin sold coding setups to engineers at big companies and made a few hundred thousand dollars a year.
In 2025 he rebuilt his company Daytona for AI agents and started over at zero.
Daytona reached a pace of $1M a year in under 50 days. Most customers sign up on their own.
Daytona has grown about 51% a month since then.
Will Bryk built a search engine for machines called Exa. It charges for each search, so it earns more each time an agent runs one. In about 3 years Exa grew to $100M a year.
A typical business software company takes about 9 years to get there.
Investors call it net negative churn when customers who stay spend enough extra to outweigh the money lost to the ones who leave. Since an agent pays more the more it works, charging by use makes net negative churn possible.
If your customer's agent does twice the work next month, you get paid twice as much without selling anything.
Guillermo Rauch runs a website hosting company called Vercel. Researchers asked the coding agent Claude Code to put a JavaScript app online. It picked Vercel each time.
Coding agents went from under 3% of Vercel's new business at the start of 2026 to about half by October.
Zeno Rocha built an email service for developers called Resend. In the same tests Claude Code picked Resend about 9 times as often as the older market leader SendGrid.
An agent tends to pick the same tool each time, so the top pick for a job gets most of its customers.
At the search tool company Ahrefs, AI models sent 0.5% of visitors and 12.1% of signups.
Agents are a new kind of customer. They multiply by the million each day, work while their owners sleep and pay for each job they run.
What could an agent buy from you?