Rebuilding IT From the Ground Up for the AI Age: Serval's Jake Stauch
In a Nutshell
Jake Stauch founded Serval to replace ServiceNow with an AI-native platform that lets users describe desired workflows and databases in natural language, which the system instantly converts into production code. Serval uses a two-agent architecture—an admin agent that builds approved tools and a help desk agent restricted to those tools—to deliver secure, automated employee support while preventing uncontrolled duplication. The company runs an AI-first organization that eliminates entire roles like solutions engineering and SDRs, prioritizing extreme talent density and rapid self-disruption over scale.
These notes were generated by AI and may contain inaccuracies.
There's always a gap between the idealized vision of what you think your job is going to be and what your job actually is. This is true for every profession. Kids idealize roles like firefighter or astronaut, and then when they get into the job they realize there's a lot they don't like. Serval wants to be the tool that closes the gap between what you think your job is going to be and what your job actually is.
Jake Stauch is the founder and CEO of Serval. The company is building the next generation ServiceNow, described as an AI-native service now. Serval is a platform for employee support. The technical term is enterprise service management, which means getting help at work. The ideal way to get help at work is to ask for something and get it instantly and automatically without waiting for someone to track down and assign a ticket.
Employees need help at work. The goal is to provide automated support so people get help immediately rather than waiting on manual ticket assignment. This requires some kind of automated support, and that automation is best built with AI. The approach is to think from first principles about how to support employees by automating all requests using AI.
When ServiceNow was founded around 2007-2008, it represented a big step function change over Peregrine and Remedy. The key insight was to think about enterprise software as workflows on top of a database. ServiceNow built a flexible workflow configuration engine on top of a database, which was groundbreaking at the time. However, the problem with their workflows and databases is that they require a lot of manual effort to build and maintain. Building those workflows often requires dedicated development resources and can take weeks to months.
In an era where business processes are changing very rapidly, by the time a workflow is implemented the business may have already moved on and want a different workflow. Automation runs behind where it should be. The same issue exists with databases. Manually updating IT assets and keeping them current is very painful and requires consultants or internal developers.
Serval keeps the core primitives of workflows on top of databases but allows AI to build the workflows and update the databases. The codegen engine lets users describe the workflow they want including all steps, permissions, approvals, and logic. The system turns natural language descriptions into code so workflows appear instantaneously with practically zero development time. The same approach applies to databases: users describe what data they want from which sources and the system generates the code to fetch and keep that data up to date without manual intervention.
The insight is that to drive enterprise automation, you have to make the process of building the automation just as simple if not simpler than the workflow being automated. When presented with a task like resetting a password, people have two options: go into Google Workspace and do it manually, or open a workflow builder and drag triggers and responses to build a custom workflow. Most people choose the manual option because building automation is harder. If building automation were actually easier, people would choose automation.
When automation becomes very easy to create, duplicate workflows become a real problem. Someone might build the 20th password reset workflow when 19 nearly identical ones already exist. This creates confusion for the AI about which workflow to run. Serval built a new agent on top of the platform that has full contextual awareness of all workflows ever built, how they work, and what they do. When a user requests a workflow that does X, Y, and Z, the agent points out the existing duplicates, suggests deleting redundant ones, separating remaining workflows into categories, and adding appropriate approval steps. The agent acts as an expert assistant that translates business requirements into actual product implementation.
Jake Stauch is known for being extremely customer focused. Rather than relying on a single north star metric, he practices full immersion by being in every single customer Slack channel and sending messages daily. This approach gives him a direct pulse on what is working and what is not. With over 100 customers including many large enterprises, staying embedded in these conversations is time-intensive but provides irreplaceable insight. Customer insight and empathy become a key moat because product advantages can be copied overnight.
After ChatGPT launched in November 2022, many people derided application layer companies as wrappers on top of foundation models. Serval's view is that the company must be happy when new models come out. The guiding principle is to build something that is not made obsolete by new models from labs and hyperscalers but is actually made better. The product provides boundaries and controls that limit model capabilities in enterprise environments. The real limitation is not model capability but how to deploy models company-wide in large enterprises without elevating security risk.
Serval divides agents into two parts. The admin agent helps build tools and skills that configure what the help desk agent can do and knows about. The help desk agent is what end users talk to when they have problems. The help desk agent can only use tools and skills that have been expressly built, published with approvals and permissions by the admins. This architecture allows the help desk agent to run with full reasoning ability and intelligence while being restricted to only the tools the IT admin has approved, some of which may have additional approval gates.
Serval uses OpenAI and Anthropic models and continuously experiments with the latest models from all providers. Different models perform better at different tasks. For end-user interaction, OpenAI models have shown the most consistent success for calling correct tools and responding appropriately. For automation and codegen, Anthropic models perform best. New model releases are not always plug-and-play. Some things improve while others that were working well may degrade because of prompt tuning and infrastructure built around previous model quirks. The team runs evaluations and does slow releases across customers. In some cases, they have downgraded back to older models when new ones introduced unpredictable behavior despite being smarter.
Cost has not been a primary focus to date because the product needs to feel magical first. Serval's unit economics are stronger than many AI companies because they are not reselling tokens. Automations are built once as TypeScript code and then run without regenerating code each time. A growing library of automations reduces the need to generate new code for common requests. The team has been told to spend more money and use the best possible models because long-term optimizations are known to exist. Cost becomes more relevant when exploring background agents and long-running agents that investigate historical tickets and device logs, as those workloads could run away quickly.
If OpenAI or Anthropic decided to build their own enterprise service management product, the competitive dynamics would be challenging but not necessarily fatal. History shows smaller companies can beat incumbents with infinite resources. The key issue is focus. Anthropic has recently added more ARR in a couple of months than ServiceNow has added in 20 years. It would not make sense for these companies to divert their best talent to a category that would take years to master when they can generate far more value from their existing product portfolios in months. They might build simple mid-market or SMB versions, but mastering the complexities of true enterprise service management would require focus that would be a poor use of resources.
The pain points and problems are remarkably similar between AI-native companies and large enterprises. Serval works with companies ranging from a few hundred employees to a few hundred thousand employees. The main difference is decision-making speed and coordination. At smaller companies, an IT leader can decide how processes work. At large enterprises, decisions require committees and coordination across many stakeholders. This makes implementation and adoption slower. In AI-native companies, passionate IT people get to spend time building and using modern tools rather than managing legacy ticketing systems. In large enterprises, the impact is broader because employee experience problems are deeper. People can be blocked for weeks waiting on responses and multiple tickets create confusion.
Serval forces the team to use the product for everything, including office requests like snacks. One notable internal use is the Dream Team Draft channel. The company takes a person-first approach to recruiting by identifying the best people in the world rather than casting a wide funnel. People post the best people they have ever worked with in this channel. Serval then takes that profile and runs a series of automations that integrate with outbound campaigns.
Serval runs outbounding campaigns and nurturing campaigns that warm up potential talent. The system ensures candidates see Serval everywhere, making the company top of mind. Since these individuals are generally not leaving their current roles immediately, the approach plays the long game. Employees can simply say they love a candidate, and the talent team uses Serval's automations to bring them into the system.
The greatest reason people join is the team itself. Candidates consistently remark that the team is kind, talented, and fun to be around. The energy in the office is contagious. While there are interesting technical problems—building complex enterprise automations and touching HR, legal, finance, IT, and security—the primary draw is meeting the team and realizing it is the right place.
Serval is not a place with extensive training or mentorship. Everyone is kind, but people are expected to be productive on day one. There is no structured onboarding program, coaching, or paired resources. The company hires people who thrive in ambiguity and figure out what needs to be done. There are no clearly defined career paths or clear progression up an org chart. The organization stays as flat as possible, and the founder remains blissfully unaware of formal reporting lines.
Beyond using Serval internally, the company questions every role and department, starting with the assumption that a function might no longer need a human. AI gets the right of first refusal for every job. Solutions engineering has been fully replaced by AI. There are no solutions engineers and no SDRs. Sales reps use Serval to answer product questions instantly, generate decks, one-pagers, battle cards, and comparison sheets on the fly during calls. Four deployed engineers assist with pilot implementations.
The company delayed hiring in many areas and discovered how far AI tools can scale enablement and other functions. Product marketing has built automated enablement resources. RevOps reached a point where some human involvement became necessary, but the department ended up smaller than originally planned.
In the earliest days, the founder felt useless while the CTO built the product and the team struggled to generate customer interest. The focus was on hiring, setting up calls, and trying to sell. Over time the dynamic shifted from pushing the business forward to the business pulling the founder along.
The founder remains heavily involved with customers on both the sales side and long-term success conversations, as well as recruiting. Involvement in product direction has decreased. Product improvements now emerge through the four deployed engineers who receive direct customer feedback and rapidly iterate, creating a closed loop that feels like gradient descent for product improvements.
An AI-native organization must be willing to reinvent itself and disrupt itself much faster than traditional companies. Parts of the product will be renamed, and entire approaches will shift. The mindset has moved from building software that lasts twenty years to building software that may only last six months before requiring a rebuild when paradigms change.
Serval sits between rapidly evolving foundation models and customers whose needs change more slowly. The company serves as a translation layer. Deployed engineers start with customers' actual business problems and then show how those problems are solved in Serval. At the same time, the team incorporates the latest AI advances to deliver those solutions. Customers are educated on new capabilities and how approaches have changed since three months prior. Work is underway to let end users simply state their business problems while Serval handles the underlying AI implementation.
A significant tension exists between individuals who want autonomy for their agents and organizations that want control. Employees desire agents with broad access and freedom to act. Security and IT organizations are concerned about the risks. This mirrors the earlier shadow IT phenomenon when employees adopted iPhones despite enterprise preferences for locked-down BlackBerrys. Companies that default to saying yes to AI adoption will likely pull ahead, even while accepting that security incidents will occur and valuable lessons will be learned from them.
Hiring remains the top priority and concern despite AI's supposed ability to reduce headcount. People are still the strongest moat. The company follows a "fewer, better" mantra, repeatedly asking how to make each department smaller while raising quality. High talent density enables the agility required to reinvent the organization frequently. A small, high-caliber team can shift direction almost overnight, whereas scaling to thousands of employees would make rapid change far more difficult.
If Serval becomes a large, successful company, the founder hopes the clearest impact will be that the product unlocked meaningful work for people. Rather than being remembered simply for automating IT jobs, the goal is for Serval to close the gap between the idealized vision of a job and the actual day-to-day reality by removing repetitive, menial tasks that people did not sign up to do. The aspiration is that employees across organizations feel their work lives became significantly better because of Serval.
Keep Sequoia Capital in your library
Save the videos and channels worth coming back to, and find them again in one place.
