Leadership1 publisher2 min readPublished
Agents delivered a million signups to a product almost no human had heard of
A startup CEO told Business Insider the signups arrived in a week or two with no launch and no marketing. Cloudflare estimates bot traffic already exceeds human traffic and could reach 1,000 times it within five years.
The Board Room · Leadership desk
What happened
- A startup CEO told Business Insider that a product his company put online with no launch and no marketing drew at least a million signups in a week or two, almost all of them from AI agents.
- Cloudflare estimates that bot traffic overtook human traffic a few months ago and could run 1,000 times larger than human-generated traffic in about five years.
- Browser Use cofounder Gregor Zunic posted a video of an agent finding the best Zurich-to-London flight in seven seconds at a cost of $0.0039 for the task.
Compiled by The Board RoomSomething wrong?How this is made
Why it matters
- decision Pricing built for human users, per seat and per month, does not fit a caller that can issue thousands of requests a day for fractions of a cent each, so the pricing model comes up for review sooner than the traffic does.
- exposure Incumbents whose margins depend on customers not shopping around are now reachable through software that does the shopping and the cancelling for a customer who never called.
- contradiction The evidence that agents are an acquisition channel is one unnamed company's signup story, while the evidence that they are a load problem comes with a named company's estimate and a shipped product, so one half of the case is far better sourced than the other.
Signups are the cheapest number in software, and this one carries the thinnest sourcing in the story. Business Insider did not name the company, the product, or any revenue attached to the million [1][2]. Spread over two weeks, a million accounts is about 71,000 a day; over one week, about 143,000 [4].
The cost of producing volume like that is visible. Gregor Zunic, cofounder of Browser Use, posted a video on X on September 17 showing an agent use Google to find the best flight from Zurich to London on a set date, in seven seconds, at a cost of $0.0039 [6]. One agent repeating that task without pause would issue roughly 12,300 requests a day [2]. A million such requests cost the requester about $3,900 [1]. The site on the other end pays in origin compute and in handling whatever the agent starts.
Cloudflare's projection is a multi-year number. The company estimates bot traffic passed human traffic a few months ago and could run 1,000 times larger than human-generated traffic in about five years [4]. "Humans will be a rounding error on the internet, not because human traffic goes down, but that's just how fast we're seeing non-human traffic grow," Cloudflare CFO Thomas Seifert said earlier this year [5]. Read literally from a starting point of parity, that implies the bot-to-human ratio roughly quadrupling every year for five years [3].
The capacity question does not rest on the anecdote. Muse launched last week, has been calling customer service on behalf of millions of users, and hit the top of the app charts late Thursday [3]. Support queues are staffed against forecast human contact volume.
On the buying side, the individual accounts are more specific than the traffic estimates. Joseph Devoy, founder of the startup Tempo, wrote on X: "Yesterday I uploaded my auto insurance policy to Meta @Muse and asked for a better rate with identical coverage. In ~5 minutes, it found a policy saving me $3,500 /year, bought it, and canceled my old one." [7] Business Insider's Pranav Dixit tested Instinct, an agent that lives in text messages, and within a few days it booked travel, ordered groceries, dealt with customer service, canceled subscriptions, answered emails and faxed paperwork to his bank [8].
The choice this quarter is between blocking and metering, and neither is free. Blocking non-human traffic protects the capacity plan and gives up whatever share of agent-driven signups turns out to be real. Metering means identifying agents and then rate-limiting or charging them, which is authentication and billing work set against a revenue case that so far consists of one story about signups [1].
What to watch
- Whether Meta publishes Muse call or transaction volumes per merchant, which would give operators a number to staff against.
- Whether any company identifies itself as agent-acquired and reports revenue, not signups, from the channel.
- Whether Cloudflare updates the bot-to-human ratio with measured data, since the 1,000x figure is an estimate.