Serval is an AI-native employee support platform that automates IT, HR, legal, and other internal help requests end-to-end, and is one of the fastest-growing enterprise AI applications today, recently valued at $1 billion after a Sequoia round and ranked #1 mid-stage company on the Enterprise 30.
Founder and CEO Jake Stout explains why Serval chose to build its own full system of record rather than layering on top of incumbents like ServiceNow, how it uses code generation as the backbone for workflow automation, why ITSM is uniquely vulnerable to AI-native disruption, and how the company recruits and structures teams for an AI-native world.
Owning the System of Record
Serval’s core strategic decision was to build a full platform and system of record, not just a layer on top of existing tools like ServiceNow or Jira Service Management.
Product rationale: Being beholden to a legacy platform limits how good your product can be. If the first step in a demo is configuring ServiceNow, you’ve already lost the ability to wow the customer.
Go-to-market rationale: It’s far easier to sell into an existing budget category than to create a new one. Every company already spends money on ITSM, so positioning as a “better version” of something they already buy is more reliable than asking for experimental AI budget.
The tension: Building a full system of record means the product must be extremely mature before the first sale, because enterprise buyers have high expectations. Serval had no customers at the one-year mark, which created intense doubt.
The bridge strategy: Serval builds a mirror of the customer’s existing system of record, syncing all data from ServiceNow or Jira into Serval. This lets them add value immediately while making the eventual migration nearly seamless.
Why ITSM Is Particularly Vulnerable
Not all horizontal systems of record are equally susceptible to AI-native disruption. ITSM has specific characteristics that make it more vulnerable than CRM, ERP, or HRIS.
Lower data urgency: ITSM data (e.g., a password reset request from last week) is less critical to daily workflows than CRM or ERP data, which people need at their fingertips constantly. This makes organizations more willing to migrate.
Faster-changing workflows: IT tools and applications change rapidly, meaning the workflows ITSM must support shift faster than those in ERP or CRM, where processes are more stable.
Higher switching frequency: Companies already change ITSM platforms more often than other systems of record, and AI is accelerating this trend.
Legacy incumbents are behind: ServiceNow and Jira have not delivered compelling AI products, creating an opening. Large enterprises are now actively exploring alternatives rather than waiting.
Automation via Code Generation
Serval’s key technical bet was that automation workflows should be built in code, with code as the source of truth, rather than using traditional drag-and-drop workflow builders.
The insight: People can describe a complex workflow in a single sentence but spend days or weeks building it node-by-node in a visual tool. Code is naturally suited to turning a concise description into a deterministic flow.
The bet (made ~2 years ago): At the time, code generation was unreliable and this approach didn’t work well. But the team bet that coding models would improve dramatically, and that first principles favored code as the right abstraction.
How it works today: Users describe any automation in natural language (onboarding, offboarding, reporting, password resets, etc.). Serval generates the underlying code. No technical skill is required because the interface still looks like a visual workflow builder; code is the source of truth behind the scenes.
Model improvements have been transformative: The biggest leap came from improvements in coding models (especially around late 2024). What used to require heavy prompting and guardrails now works reliably for virtually any request. The help-desk agent side (understanding user questions, routing to tools) has been more stable and improved less dramatically.
Critical Guardrails for Enterprise-Grade AI
Running user-generated code against mission-critical business systems requires extensive safeguards.
Architecture as guardrail: The agent that interacts with end users in Slack or Teams does not have direct access to business systems (Google, Microsoft, Workday, ServiceNow). Instead, it routes requests to deterministic workflows that IT and security teams have explicitly built, published, and approved with defined permissions. This makes it physically impossible for a user to trick the agent into unauthorized actions.
Code validation: Generated code is type-checked, compared against ingested API specs from connected applications, and validated against provided examples.
Visualization as a check: Serval visualizes what the code does by having an AI describe the code’s behavior. If the code hallucinates or inserts an unwanted step, it shows up in the visualization, so non-technical users can catch problems without reading code.
Eval suite built from failures: Every failure case is added to an evaluation suite. Nothing ships unless it passes all previous failure cases (100% pass rule). This prevents regressions when new models are tested.
The Future: Employee and Agent Support
Serval’s long-term vision extends beyond human employees to support AI agents as well.
Agents will need to make access requests, go through approval procedures, and have those requests logged, just like humans. Serval is building the infrastructure for both people and agents to have requests resolved through a single system.
The theory is that agents should behave somewhat anthropomorphously in requesting access, because it’s easier for humans to reason about and trust a unified system than a separate agent-only permissioning model.
Near-term unlock: Computer use and browser automation for the long tail of tasks not solvable via APIs or MCP, especially for legacy enterprise tools with poor API coverage. Key remaining challenges include reliability, authentication, security, and determinism when interfaces change.
Competing with ServiceNow
Competing against ServiceNow (~$14B ARR) is “incredibly difficult” — they are formidable, well-resourced, and the incumbent for good reason.
Startup advantage is velocity: The only real edge is speed. Serval needs to ship product before a legacy incumbent can even hold a meeting about shipping product.
Talent density matters: Small teams of exceptional talent can take down large enterprises. This compounds over time.
Serval’s defensibility vs. foundation model risk: Unlike AI companies whose value proposition disappears if you turn AI off, Serval has deep application value independent of AI (workflows, permissions, logging, integrations). This makes it less vulnerable to being absorbed by foundation models like ChatGPT.
Horizontal SaaS consolidation: Jake believes the future will have fewer but larger horizontal SaaS companies clustered around major systems of record, with increasing overlap between them. Point solutions will struggle to survive as standalone companies.
Internal Support vs. Customer Support
People often compare ITSM to customer support as AI-native verticals, but the differences are significant.
Volume: Employees ask for help at work constantly, far more often than customers call support lines. Internal support is the larger category.
Complexity: Internal support requires deep integration with privileged business systems (IDP, AWS, GitHub, Workday) with strict security and permissioning models. Customer support tools don’t face this.
User attitude: Employees contacting internal support are generally in a better mood than angry customers, which affects product design — users are willing to take more turns with an agent and do more triage.
Hiring as the Only Durable Moat
Jake believes talent is the only durable competitive advantage, because nearly everything else is copyable within 6-12 months by a motivated team with modern AI tools.
“Dream Team Draft”: Every employee’s #1 job is to recruit the most talented people they’ve ever worked with. A Slack channel called “dream team draft” lets anyone post LinkedIn profiles of exceptional people they know. Serval workflows then take those profiles and execute a targeted recruiting plan, similar to how professional sports teams recruit specific athletes rather than holding open tryouts.
Person-first recruiting: Instead of mass campaigns and funnels, Serval identifies specific individuals, builds a plan to recruit them, and tailors the pitch. This is inspired by the Ocean’s 11 approach of recruiting specific people for specific skills.
Junior engineers are a strength: Junior engineers at Serval are highly AI-native, giving AI “right of first refusal” on every task. Senior engineers tend to be more hesitant, having established workflows and opinions about the right approach. Having both on a team creates a productive tension.
Team Structure and Management in the AI Era
The role of managers is in flux due to competing forces.
ICs own more scope than ever: AI tools let individual contributors build huge swaths of product surface area, potentially reducing the need for management and coaching.
But tracking becomes harder: When each IC is producing vast amounts of work (including AI-generated code), it becomes more difficult for managers to review, provide feedback, and keep tabs on everything.
Trend toward flatter orgs and super ICs: The fastest-moving companies tend to have smaller teams with more autonomy. There’s growing skepticism of pure managers; the preference is for “super ICs” who may have people working with them but aren’t building empires.
Bias toward senior talent: There’s a trend toward hiring more senior people on the tech side, raising questions about how junior engineers will develop if most resources go to senior staff engineers.
Forward Deployed Engineers
Serval’s version of forward deployed engineers differs from the typical implementation or solutions engineer role.
They are software engineers who build the platform while embedded with customers in a constant feedback loop.
Why this matters: Legacy incumbents like ServiceNow have enormous product surface area built over 20 years. Each enterprise customer uses only about 5% of that capability, but every customer uses a different 5%. By embedding engineers with customers, Serval identifies the specific product gaps preventing wins and closes them one by one.
Jake describes this as “gradient descent for product building” — iteratively encountering gaps and improving the product toward a global maximum.
Go-to-Market: Building for Large Enterprise from Day One
The ITSM market is overwhelmingly concentrated in large enterprise (90%+ of ServiceNow’s revenue comes from $1M+ ACV customers).
Rather than starting small and moving upmarket (the standard startup playbook), Serval always built with large enterprise in mind because that’s where the dollars are and where a major horizontal software company must play.
Early customers were smaller fast-growing companies (Perplexity, Mercury, Clay, Together AI) because large enterprises won’t buy from an early-stage startup. But now Serval is shifting upmarket, having closed its first Fortune 20 account, with more in pipeline.
The ICP is essentially “every company in the world” — global GDP.
Quickfire
Changed mind on: How big this could be. A year ago the goal was “don’t die.” After a week of conversations with 5-10 Fortune 100 companies, all excited about the product, Jake realized this could be one of the biggest software companies in history.
Most excited about: Computer use and browser automation unlocking the long tail of automations not solvable via APIs, especially for legacy enterprise tools with poor API coverage.
Overhyped: Weekend POCs that automate “fake jobs” — there’s a big gap between a cool demo and an enterprise-grade workflow that a large company trusts in production.
Underhyped: Home robotics — Jake believes they’ll be the next automobile in terms of economic impact and household adoption.
Infrastructure with staying power: Major cloud providers (AWS, GCP, Azure) are durable. Temporal (workflow orchestration) is a meaningful tool for Serval. But everything else feels vulnerable — the company has implemented and ripped out tools within two months, and increasingly builds rather than buys point solutions.
Hardest decision as CEO: Not changing course at the one-year mark when there was no evidence the strategy was working, despite pressure to pivot downmarket, build a shallow layer, or expand categories.
Background in neurotech: Jake’s first company built EEG headsets and brain-controlled video games for kids with ADHD. He’s skeptical of non-invasive EEG (too noisy, like holding a microphone outside a soccer stadium) and most excited about invasive neural implants, which would provide far richer signal.
