Q&A: Serval's CEO explains how they are building an 'AI-native' alternative to ServiceNow

Serval, an IT service management (ITSM) startup launched two years ago, positions itself as an "AI-native" alternative to established vendors such as ServiceNow and Freshworks. According to Jake Stauch, the company's CEO and co-founder, the core difference is not that Serval has AI and traditional vendors do not, but that its platform was built from day one around AI, without the burden of two decades of custom tables, business rules and workflow logic designed for a pre-AI world. That accumulated complexity, Stauch argues, is precisely what makes even the simplest automation changes slow and costly on legacy platforms, and layering AI on top does not remove the underlying problem.

Serval's value proposition revolves around resolving employee requests rather than simply tracking tickets. Stauch notes that traditional ITSM tools are good at recording that work exists, but that resolves nothing for the employee; someone has to manually build all the automations the workforce needs. Historically those automations have been created as an add-on, through drag-and-drop workflow builders that take months to develop and require constant maintenance whenever a business process changes, for example when adding a new approval step for password resets. Serval instead bets on letting AI do what it does best: write code. Its code-generation agent builds workflows for both simple help desk requests and longer processes, including employee onboarding and offboarding, role changes and reporting requirements.

Serval's architecture is described, in simplified terms, as two separate agents. On one side, an employee-facing help desk agent that can resolve requests phrased in natural language via Slack, Teams, phone or email (for example, "I'm locked out of my computer" or "what's the wifi password?"). On the other, an agent called Catalyst, aimed at administrators, that builds those automations and helps deploy them. Catalyst, made generally available the week before the interview, extends the capabilities of Serval's earlier Workflow Builder: while the latter translated a single instruction into a code-based workflow, Catalyst can generate multiple related workflows at once — for example, password resets for different identity providers such as Okta, Google and Microsoft — while being aware of what already exists. It can also analyze data from systems such as ServiceNow to build dashboards, propose automations based on that analysis, and even generate user interfaces, such as a form for requesting equipment purchases.

One of Catalyst's new features is the ability to create long-running background agents that detect problems from ticket history and other data, proposing fixes before employees ever experience the issue. The use cases mentioned include identifying security vulnerabilities, reclaiming unused software licenses and repairing workflows that have failed.

On the risk of these AI agents acting unexpectedly, Stauch explains that the architecture rests on a separation of powers: one agent builds the automations and a different one uses them. Catalyst is restricted to administrators only and generates deterministic, code-based workflows; the help desk agent that interacts with end users can only invoke workflows already published by administrators, with no ability to create new ones or step outside that pre-approved scope. Each workflow carries its own permissions and approval points that do not depend on the language model's judgment, so any action by the help desk agent can be traced back to a specific workflow run, and auditors can review both the approvals and the underlying code. For Stauch, that combination of deterministic tools and an "air gap" between the agent that builds and the agent that uses the tools is what allows IT teams to benefit from AI without handing broad system access to a model making real-time decisions.

As for why incumbent vendors such as ServiceNow could not simply replicate this approach, Stauch clarifies that the obstacle is not that they cannot ship agent-building features — several already have — but that their existing customers depend on years of customization built on the old architecture, and those vendors cannot retire it without breaking what those customers already use. Serval, by contrast, built the automation engine and the system of record together from the start, with no legacy structure, so creating a workflow, applying governance and capturing the data that improves future automations all happen in one place. According to Stauch, this is why many customers choose to replace their previous ITSM tool entirely rather than use Serval as an added layer: when Serval is the system of record and not a layer on top of another one, every ticket and workflow feeds directly into better automation suggestions over time.

That said, Serval can initially be deployed on top of ServiceNow or another existing ITSM tool, as a starting point for customers who spent years implementing their legacy system and are not willing to abandon it immediately. In that scenario, Serval syncs data back to ServiceNow while acting as the automation and orchestration layer, and at the same time works as its own system of record, generating tickets and building its own configuration management database. That means, according to Stauch, temporarily having two systems doing the same thing, one of which also automates the work and is not twenty years old. Some customers migrate the entire platform right away if their ServiceNow contract renewal is approaching; others, with more distant renewals, take a more gradual approach.

On Serval's reach beyond IT, Stauch says no customer uses the platform exclusively for that department: IT is actually a minority of the teams using Serval today, even though it is the typical buyer for this product category because it is the one that brings it into the organization, from where it spreads to human resources, finance and legal. According to him, a typical Serval customer has 13 different departments deployed, including IT, security, finance, legal, sales and operations.

On the impact of automation on IT teams themselves, Stauch argues that for many companies the IT team is the most technical — and often most expensive — resource, and yet spends its day resetting passwords and moving users between groups, which he considers a waste of their talent. When administrators stop doing that work, Stauch says, they can be redeployed to other functions in the organization, acting as internal engineers who build automations for the rest of the business, thereby turning IT from a service function into a building function. Asked directly whether automation shrinks IT headcount, Stauch responds that what they consistently see is redistribution, not reduction: automation frees up administrators' time — in the case of Perplexity, a Serval customer, between one and two hours a day per administrator — and that time is reinvested in building automations for human resources, finance, legal, security and engineering. In many cases, he notes, it is less about IT shrinking and more about IT managing to keep pace with headcount growth across the rest of the company without having to add staff of its own, although administrators still verify workflows and approve those that touch sensitive systems.

On the corporate front, the article notes that Serval counts Spotify, Fox and Live Nation among its customers, and that it has raised significant venture capital funding from Sequoia and other investors, at a most recent valuation of approximately $1 billion.