Jake Stauch: How Serval Takes On ServiceNow With Code Agents

How Serval Brings Coding Agents to IT Teams to Replace ServiceNow

Jake Stauch is the co-founder and CEO of Serval, an enterprise service management and automation platform that competes head-on with ServiceNow. Before Serval he spent years at Verkada selling physical security products to IT teams, and he co-founded Serval in 2024 with Alex McLeod, another Verkada alum. On No Priors he explains why Serval went straight after the biggest incumbent instead of building a layer on top of it, how its "coding agents for the rest of us" stay safe inside large enterprises, and what running a company that grew tenfold in a year looks like.

Code is the automation layer: Stauch's founding bet was that ServiceNow is, at its core, a workflow automation platform, and that code beats drag-and-drop builders for expressing automation. Drag-and-drop, he says, is too complex for non-technical users and annoyingly abstracted for technical ones. Code only lost because few people could write it. Coding agents changed that: password resets, approvals and offboarding are deterministic, verifiable scripts, the kind of work he expected models to get good at. Early demos seeded the model with a library of example workflows. Then the models got good enough that it became hard to give Serval a workflow it couldn't automate, as long as the APIs existed.

IT is the engineering team: "Most companies have an IT department, not a team of software engineers." Serval gives IT the same safety net engineers rely on: generated code goes through staging, approval and an audit trail before it reaches production, and IT then extends the platform to HR, finance, people and legal teams. Stauch says Serval treats IT "like an adult": the code is always visible for an IT engineer who wants it, while a marketing or people-team user never has to see it. An agent called Catalyst lets admins describe what they want and have it translated into automations, approval rules, permission checks and policies, so customers don't have to learn the product's primitives first.

Guardrails are the product: "Serval is the boundary. Serval is the guardrails." Stauch argues that personal agents like OpenAI's Dot work for consumers and smaller companies but hit a wall in large enterprises, where an individual's permissions barely let an agent read email. Serval instead connects with service-account and admin credentials. When a user wants an agent to do something privileged, Serval opens an access request, routes it through approval, grants just-in-time access to run a deterministic workflow once, and logs it all in its own ticketing system. He expects enterprises to shrink everyone's permissions to strict least privilege as agents raise the damage any one account can do, and sees that as Serval's advantage: "by owning how privileges get escalated in the enterprise, we actually become more powerful and more essential the more you clamp down on end user permissions." Background agents follow the same pattern: they ask for new deterministic tools, a human approves each one, and the agent's toolkit grows over time.

Forward deployed, then ProServal: Serval sends a deployment strategist and a deployed software engineer on site with customers to map processes and implement them. The rule is that they never build custom features. Per-customer differences live in the generated automations, and when many customers build the same automation by hand, it becomes a product feature. Two examples: detecting conflicting or outdated knowledge base articles, and reclaiming unused software licenses. Because demand kept growing after go-live, Serval added an automation engineering team it calls ProServal for ongoing work.

Pricing per workflow, and the SaaSpocalypse: Stauch says CIOs are pulling back from early token maxing but still plan to increase spend. What they need is proof of ROI. "We try to price on workflows run because I think it's a better reflection of actual work being done versus just tokens", which he says reflect verbosity and inefficiency more than work. He doesn't expect AI to wipe out systems of record: companies that tried to vibe-code their core tools found enterprise software expensive to build and maintain. The consolidation he expects hits point solutions and middleware, while ERP, HRIS, ITSM and CRM survive under a new generation of incumbents. His proof point on sales: Fox Corporation adopted Serval company-wide, and the person who led the migration was nominated as the company's MVP.

20 to 200 people, no product team: Serval is two and a half years old and had 200 people at recording, up from 20 in January. Stauch plans to hire even faster, arguing that because each employee does more, the company wants more of them, not fewer. Engineers sit in customer Slack channels and most join at least one customer call a day, so product decisions are pushed down to them: "We don't have a product function."

5 Takeaways from Jake Stauch on Enterprise Coding Agents

• Attack the incumbent directly - Serval skipped both the "AI layer on top of ServiceNow" and the down-market playbooks and built for large enterprises from day one
• Deterministic code beats free-roaming agents - Generated workflows run through dev, staging and approval, which is what regulated companies need before they ship agent capabilities
• Least privilege plus approvals - Just-in-time, audited privilege escalation lets agents do admin-level work without giving every employee broad access
• Turn repeated automations into features - Deployed engineers never build bespoke features; patterns many customers build by hand become product
• Price on outcomes, not tokens - Charging per workflow run gives buyers the ROI proof they now demand after the token-maxing phase

What Serval's Approach Means for AI Agents in the Enterprise

Stauch's argument is that enterprise agent adoption is limited by permissions, not intelligence: smarter models make companies tighten access, not loosen it. The winners will own the approval, audit and escalation layer around agents, and will run it like a software development lifecycle. For any organization deploying AI agents, the practical lesson is to keep agent actions deterministic, approved and logged, and to measure value in workflows completed rather than tokens burned.