Some startups take pride in their high-profile financial backers, but equally vital is having prominent customers.
This is a key point of pride for Serval, an enterprise AI firm that revealed a $47 million Series A funding round on Tuesday. Redpoint Ventures led the round, with involvement from major venture capital firms like First Round, General Catalyst, and Box Group. However, the company’s client list, which includes significant AI players like Perplexity, Mercor, and Together AI, is even more remarkable than the investors.
In general terms, Serval is employing agentic AI models to automate IT service management, but the company’s distinctive strategy leverages agentic AI’s capabilities while mitigating its risks. One agent is dedicated to coding internal automations for routine tasks, such as authorizing software or provisioning a device. The founders envision it as a type of vibe-coding tool, supervised by an IT manager, yet independently handling most of the work. A separate help desk agent addresses user inquiries by executing these tools as directed, adhering to the tool’s established rules.
Serval CEO Jake Stauch emphasizes that the simplification of the tool-building process was crucial.
“We want to eliminate any sense of marginal cost associated with building these automations,” Stauch stated to TechCrunch. “Our goal is to make automating something permanently easier than performing it manually once.”
Dividing the task between two agents—one for tool creation and another for tool utilization—also provides managers with a mechanism to monitor permissions. Upon the creation of an automation, the manager establishes usage rules, offering an additional safeguard against overly enthusiastic help desk agents.
Enterprise clients are acutely aware of the potential dangers of a wayward AI system, which is a factor in Serval’s decision to forgo a single, all-encompassing Help Desk Agent.
“You don’t want someone to enter Slack and request the deletion of all company data, only for the AI agent to respond helpfully with, ‘Sure, I’ll delete all the data,’” Stauch explained to TechCrunch. “Instead, it will respond with, ‘I don’t have a tool for deleting all the company data. However, I do have a tool for resetting your password or performing these other tasks.’”
Given their deterministic nature, the tools can incorporate highly intricate permissions, such as permitting actions only after a specific multi-factor authentication process or within a defined timeframe. Furthermore, an AI agent is readily available to modify the codebase whenever these rules require adjustment.
It represents a fresh perspective on the prevalent challenge of overseeing agentic AI systems. “You need complete insight into and control over the AI agent’s actions,” Stauch asserts. “And you achieve this by leveraging Serval to develop these tools and tailor the associated permissions and approvals.”
