Serval’s Jake Stauch on coding agents for enterprise IT

Serval co-founder and CEO Jake Stauch joins No Priors hosts Sarah Guo and Elad Gil to discuss enterprise automation. He argues that coding agents could help IT and other departments build workflows, but that large organizations need explicit permissions, approvals and audit trails before they can delegate sensitive work.

Stauch describes Serval as an enterprise automation and service-management platform that competes with ServiceNow. He encountered the category at Verkada, where he spoke with IT teams about what they spent money on and what frustrated them. ServiceNow often appeared in both answers.

He credits ServiceNow as a powerful, customizable platform but says customers told him implementing changes could be slow and expensive, sometimes requiring consultants. A long implementation may be acceptable for a stable onboarding process; it is harder when a company wants to keep adding smaller automations.

Guo and Gil ask about extending coding agents beyond software engineers. Stauch says engineers can experiment because generated code goes through review before production. Serval’s goal, he says, is to bring similar building capabilities to IT, HR, finance and other teams through abstractions, staging, approval and audit controls.

Stauch distinguishes that approach from a personal assistant with an employee’s ordinary permissions. In a large enterprise, permissions are often narrow and many systems cannot be touched freely. The constraint limits what a personal agent can do, even if the underlying model is capable.

He says Serval connects through managed accounts and structures exceptional access around explicit approval and documented workflows. An access request can be reviewed, logged and granted for a particular task. He presents this as a way to make useful automation compatible with enterprise governance, not as a magic guarantee of safety.

Customers still need people to understand their processes and decide what to automate. Stauch says Serval deploys strategists and engineers to help implement the platform, then offers ongoing automation-engineering support through ProServal for the long tail of requests.

At roughly two and a half years old, the company is taking aim at large enterprise systems directly. Stauch says that ambition existed from the beginning, despite the difficulty. The hosts contrast it with selling a small layer on top of an incumbent or starting in smaller businesses and moving upmarket later.

Stauch argues that targeting large customers from day one changes product, hiring and growth decisions. His team chose service management because it understood IT workflows and believed code was a more expressive way to implement automation than drag-and-drop builders.

The obstacle was that relatively few people could write and maintain the code. Stauch and his co-founder bet that coding agents would improve, particularly on well-defined, verifiable procedural tasks. In early demos, they supplied many examples; later model improvements reduced how much scaffolding they needed.

He welcomes further advances in coding and computer use because Serval does not train the underlying models itself. He describes its product as the boundary around those capabilities: controls that let large organizations use increasingly capable tools within their own processes.

Stauch predicts enterprises will narrow individual permissions as agents make it easier to act at scale. Exceptions will still be needed, so he sees a role for approval workflows and audit trails when access must be elevated. This is his forecast and product strategy, not a demonstrated universal outcome.

The hosts ask why a customer would risk replacing an embedded incumbent. Stauch says the company must make the change a career success for the internal champion. He cites Fox as a customer and says the person leading its migration received company recognition; the conversation does not independently evaluate the deployment.

Looking ahead, Stauch expects personal agents at work to communicate and delegate to one another. He also expects a tension between employees’ desire for convenience and organizational security. He thinks personal and work agents may remain separate, much like personal and corporate email.

Asked for product principles, he points to the software development life cycle: development and staging environments, review, testing and gates before production. Some IT systems already have such controls, but Stauch argues new AI products too often skip them.

Serval uses deployed staff as well as a core platform. Stauch says the company does not build bespoke product features for individual customers, although customers create different automations on the same primitives. Repeated customer-created workflows can reveal which capabilities deserve to become standard features.

He gives two examples: finding outdated or conflicting knowledge-base articles and reclaiming unused software licenses. When many customers repeatedly build the same pattern, Serval may incorporate it into the platform instead of asking each customer to recreate it.

The hosts turn to AI spending. Stauch says some customers are reining in an initial rush to consume tokens, while still expecting spending to grow over time. What they need, in his telling, is evidence of return on investment, which remains difficult to measure.

Serval tries to price around workflows run because Stauch sees that as closer to work performed than token volume. He expects the industry to keep experimenting with how to compare AI spending with work previously done by software or people.

On the so-called SaaS apocalypse, Stauch argues that rebuilding core enterprise systems is still expensive to create and maintain, even with code generation. He expects consolidation among smaller point solutions but thinks core categories such as ERP, HR systems, service management and CRM will remain, perhaps with new suppliers.

The hosts ask how non-engineers can build robust systems. Stauch says enterprise IT teams often have more technical ability than they are credited with. Serval exposes code and a development life cycle for those who want them, while providing abstractions for users who do not want to work directly with code.

He introduces Catalyst, an agent intended to help administrators translate requirements into platform configurations, automations and approval rules. The premise is that a user can discuss what they want to build instead of first learning every product primitive.

Stauch calls background agents an enterprise moonshot: ideally they would discover and solve problems without constant supervision. He says the field has not yet resolved the tension between that autonomy and the need to know what an agent can read or do.

His proposed approach is incremental permission escalation. An agent requests access to particular tools or workflows, people approve the requests, and the agent gains a larger authorized toolkit over time. He acknowledges the process can be cumbersome and says unrestricted code execution would introduce serious security risks.

Asked whether better models or better guardrails matter more, Stauch answers that both matter. Agents need memory and context to understand preferences, but organizations will still need to decide which actions they trust the agent to take and when it must check back.

Stauch says Serval has around 200 employees and expects hiring to continue. His argument is that higher individual productivity expands the work the company can attempt rather than eliminating its appetite for people. This is the company’s hiring outlook, not a general forecast for every workplace.

To keep up with customers, he says engineers participate directly in customer conversations and channels. Serval has decentralized much product decision-making to those engineers rather than maintaining a separate product function, according to Stauch.

That speed creates challenges: customers and leadership may struggle to absorb changes. Stauch says engineers need product judgment and the authority to make local decisions, while leaders must be willing to reverse decisions when customer needs or the market change. He credits that willingness to rethink earlier choices for keeping the company adaptable.