Some startups pride themselves on having high-profile backers, but just as important is having high-profile customers.
It’s a point of great pride for Serval, an enterprise AI company that announced a $47 million Series A round on Tuesday. The round was led by Redpoint Ventures with participation from leading venture firms including First Round, General Catalyst, and Box Group. But even more impressive than its funders is the company’s client list, which includes major AI players such as Perplexity, Mercor, and Together AI.
Broadly speaking, Serval uses agent AI models to automate IT service management, but the company has a unique approach that leverages the power of agent AI while avoiding many of the pitfalls. One agent is used to code internal automation for routine tasks such as software authentication and device provisioning. The founders see it as a kind of vibecoding tool, with an IT manager overseeing it but most of the work being done alone. Other help desk agents respond to user requests by invoking these tools in response to commands, according to rules established by the tools.
Serval CEO Jake Stauch says the key was to make the process of building the tool as simple as possible.
“We don’t want you to see a marginal cost to building this automation,” Stauch told TechCrunch. “We want to automate something much easier than doing it manually once.”
Splitting the task into two agents (one responsible for building the tool and one responsible for using it) also provides a way for administrators to monitor permissions. Once an automation is created, managers set rules for when the automation can be used. This provides an additional line of defense against overzealous help desk agents.
Enterprise customers are acutely aware of the risks of rogue AI systems. This is one of the reasons Serval decided not to employ a single all-purpose help desk agent.
“We don’t want someone to go into Slack and say, ‘I want to delete all of my company’s data,’ and a super-friendly AI agent to say, ‘Okay, I’m going to delete all of your data,'” Stauch told TechCrunch. “Instead, they’ll say, ‘We don’t have the tools to delete all your company data, but we do have tools to reset passwords and perform other tasks.'”
The tools themselves are deterministic and can include very complex permissions, such as only allowing certain actions after a certain multi-factor authentication process or within a certain time frame. And whenever you need to change those rules, an AI agent is ready to jump into your codebase and make the changes.
This is a new approach to the very common problem of how to supervise agent AI systems. “We want to have complete visibility and control over the behavior of our AI agents,” Stauch says. “And we do that by building tools using Serval and customizing the permissions and authorizations behind them.”
