One 60-person firm says its 300 agents outnumber its staff

Aiden Sochowski, a founding engineer at Runlayer, said on the AI Engineer show that agents in companies now outnumber the people running them. His evidence is his own firm: about 60 people and over 300 agents. His definition is broad, covering automated work that users do not always start, such as through a chatbot, a trigger or a software app. It is one company's count, not a survey.

What was said

Runlayer put a gateway in front of its agents. A gateway is a checkpoint that agent traffic passes through, so it can be recorded. The team logged every session and every tool call. Sochowski said this turned up several findings.

He put the scale this way: "we're a company around 50, 60 people. We have over 300 agents that are running every single day."

Not all agents are alike, he said. Some hold too many tools, which widens their blast radius: the damage if something goes wrong. He argued that agents speed up the risks of mixing different classes of data, which once took a deliberate, hand-built program.

His email example was an illustration, not a Runlayer incident. A malicious sender asks for a list of prospects, and a naive, over-permissioned agent could hand it over.

Why it matters

Our reading: the email case is prompt injection. The agent cannot reliably tell a stranger's text from its own instructions, so outside text can steer an agent that holds tools. That explanation is ours, not Sochowski's.

The thesis that follows is also ours: the unit of risk is the permission each agent holds, not the headcount. Sixty staff and 300 agents means 300 sets of tools, data reach and blast radius, which a staff list does not show. Runlayer's own route was a gateway, logging, and mapping agents by risk. Sochowski said most people at the companies Runlayer sees have zero visibility into their agents. He works for a vendor in this field, so weigh that.

The other side

This is one company, described by an engineer whose employer works on this problem. The excerpts do not say how many of the 300 agents proved risky, or whether other firms see the same ratio. The count also rests on his broad definition.

Sochowski did not link risk to agent count; he said agents differ in reach. He also doubted the habit of treating agents like people. So the headline ratio is a counting device, not a claim that agents resemble staff. He conceded limits: "Nothing is going to be 100% accurate. Nothing will be 100% effective here."