Q&A：Serval CEO谈构建ServiceNow的AI原生替代品

Since launching two years ago, Serval has positioned itself as an “AI-native” alternative to incumbent IT service management (ITSM) software vendors, promising to simplify the automation of helpdesk workflows. Serval’s ITSM platform includes a workflow builder that uses a natural language interface to build code-based workflows. While vendors such as Freshworks and ServiceNow also have workflow automation tools in their products, Serval claims its approach makes automating a recurring task quicker than completing it manually, resulting in more tickets resolved automatically. Serval — whose customers now include Spotify, Fox, and Live Nation — has received significant venture funding from Sequoia and others, and was most recently valued at around $1 billion. With its latest product update, Catalyst — made generally available last week —  Serval promises to take on more of the work needed to build and deploy workflow automations. That includes analyzing ticket histories and other data to identify processes to automate, then configuring settings and connecting integrations to help put those automations into production. Catalyst also lets admins create long-running background agents that detect problems using historical tickets and other data, proposing fixes before employees even run into an issue, according to Serval. Use case examples include identifying security vulnerabilities, reclaiming unused software licenses, and repairing failed workflows. Computerworld spoke to Serval CEO and co-founder Jake Stauch about the company’s approach to IT automation and how Serval believes it can differentiate from established ITSM vendors. The following transcript has been edited and condensed for clarity. What are the biggest challenges IT teams face when resolving employee requests? “Traditional ITSM tools are built to track tickets; Serval is built to resolve requests. These tools do a very good job of saying that you have work to do, but that doesn’t do anything for the employee. The problem with employee support is really around automation: somebody has to go and build all the automations employees need. “Historically, automations have been built in ITSMs as an afterthought, with drag-and-drop workflow builders. You build out these giant flow diagrams; they take months to build and then someone has to maintain them. But what happens when the business process changes, or you’ve got a new approval step for password resets? You’re doing surgery on all these different workflows. “Where we differentiate is making automation easier. We think that the best way to build automations is not a drag-and-drop workflow builder, but letting AI do what it does best, which is to write code. Our code generation agent builds workflows for large enterprises to automate not just simple help desk requests, such as password resets, but longer employee journeys, including onboarding and offboarding, job changes, and reporting requirements.” What are the main components of the Serval platform? And how does Catalyst fit in? “You can think of the system as two agents, though obviously it’s more complicated than this. “There’s a help desk agent for employees. An employee can say, ‘I’m locked out of my computer,’ ‘What’s our Wi-Fi password?’ or ‘My laptop stopped working,’ in Slack and Teams, over the phone, or via email — and the agent resolves it right there. “We have a separate agent, called Catalyst, that builds those automations. Admins talk to Catalyst, which can build any automation, like password reset workflows, onboarding, offboarding, reporting. Then it helps deploy them either for the help desk agent to resolve employee requests, or for others on the team to use. It even builds background agents that do investigations, pull data from different systems, and come back with reports and propose new workflows. “The old Workflow Builder was more limited, in that it translated your prompt into code and created a code-based workflow. You could say, ‘I want a workflow that resets a Microsoft Entra password,’ and it could build that. “With Catalyst, you can say, ‘Build a bunch of password-reset workflows for all the different [identity providers] that we have,’ and it’ll build workflows for Okta, Google, and Microsoft, with awareness of what’s already been built. “It can also tap into other systems to build dashboards. It can analyze my ServiceNow data and build charts showing all the info, and then propose automations based on its research. It can go a step further and build user interfaces. If you say, ‘I want a form for ordering laptops for users,’ it can build that form, which you can then deploy for equipment purchases across the company. What happens if Serval’s AI agents don’t behave as expected? What safeguards are in place to prevent them from taking unintended actions? “Serval’s architecture is built around a separation of powers: one agent builds automations, and a separate agent uses them. Catalyst is gated to admins only and writes deterministic, code-based workflows. The help desk agent that end users interact with can call only the workflows that admins have already published; it cannot create new workflows or reach outside that pre-approved scope. “Every workflow carries its own permissions and approval gates that aren’t subject to LLM judgement. This means that while the help desk agent reasons freely over the employee request, any action it takes can be traced to a specific workflow run, and auditors can review both the approvals and underlying code. “That combination — of deterministic tools plus an air gap between the agent that builds the tools and the agent that uses tools — is what lets IT teams get the benefits of AI without handing broad system access to a model making judgment calls in real time.” ServiceNow and other ITSM vendors are also adding AI agents and agent-building tools to their platforms. What is different about Serval’s approach, and what would prevent these firms from replicating it? “The difference isn’t that Serval has AI and legacy vendors don’t. It’s that Serval was built AI-native from day one, rather than layering agents onto two decades of custom tables, business rules, and workflow logic designed for a pre-AI world. That accumulated complexity is exactly what makes even simple automation changes slow and expensive on legacy platforms today; adding an AI layer on top doesn’t make it go away. “That’s the real barrier to replication. It isn’t that ServiceNow and other incumbents can’t ship agent-building features; several already have. It’s that their existing customers depend on years of customization built on the old architecture, and incumbents can’t tear that out without breaking what those customers already rely on. “Serval built the automation engine and the system of record together from the start, with no legacy structure, so building a workflow, enforcing governance, and capturing the data that improves future automations all happen in one place, instead of being stitched together across systems. “That’s also why many of our customers choose to fully replace their incumbent ITSM tool rather than run Serval as an add-on layer. When Serval is the system of record, not a layer on top of one, every ticket and workflow feeds directly back into better automation suggestions over time.” Serval can be implemented on top of ServiceNow or another ITSM tool. Can that be a long-term arrangement for customers? “It can be as a starting point – no one wants it that way [longer term]. It may have taken you six years to get ServiceNow implemented, so we don’t expect you to get off it next week. We have to give you a path to get value out of Serval on the day you start. “When you begin that way, Serval syncs data back into ServiceNow. You can continue to use ServiceNow — or whatever your legacy ITSM is — with Serval as the automation and orchestration layer on top. “While it’s doing that, Serval is also your system of record; it records requests as tickets, and builds out the configuration management database. You end up with two systems doing the same thing, one of which is also automating the work and isn’t 20 years old. “Some customers migrate the whole platform immediately, if their ServiceNow renewal is coming up. If their renewal is years into the future, they take a slower approach, using Serval for orchestration automation before the full migration.” What potential do you see for other enterprise use cases outside of IT? “We don’t have any customers that only use Serval for IT. IT actually accounts for a minority of the teams in Serval today. We talk so much about IT because IT is the buyer of this product of this category; they bring it into the organization and it spreads to other departments, to HR and finance and legal. “A typical customer of ours has 13 different departments deployed including IT, security, finance, legal, sales, operations. So, yes, we market towards IT, but IT is really the gateway into the organization.” What impact does increased automation have on customers’ IT teams? “For a lot of companies, the most technical — and often most expensive –\— resource is the IT team. Yet they spend their day resetting passwords and moving people into groups. It’s a waste of their talent, of their capabilities. “Our customers find that when admins no longer have to do that work, they can deploy them across the organization. They can bring AI to different functions, and act as internally deployed engineers that build automations for the rest of the business. You’re enabling them to be builders versus ticket takers, and transforming the IT function from a service function to a building function.” Do customers reduce IT headcount as they automate more of this work? “What we’re consistently seeing is redeployment, not reduction. Automation frees up admins’ time — in Perplexity’s case [a Serval customer], one to two hours a day per admin — and that time gets reinvested into building automations for the rest of the business: HR, finance, legal, security, and engineering. In a lot of cases, it’s less about IT shrinking and more about IT keeping pace with headcount growth everywhere else in the company without having to add staff of its own. “Automation doesn’t mean admins step away entirely. They’re still verifying workflows and approving the ones that touch sensitive systems. What changes is that they’re doing that instead of manually resetting a password or provisioning a Slack channel for the hundredth time. “What we’re seeing is an unlimited demand for the skills that IT possesses, and so when tickets get automated and IT teams can build amazing tools in Serval, they become even more valuable to their organization, not less.” IT OperationsGenerative AIArtificial Intelligence