Here is a fact about corporate life that is both boring and, on reflection, slightly insane: most companies own automation tools, pay for automation tools, and still do almost everything by hand. Someone emails IT. A human reads it. A human clicks around in five systems. A human closes the ticket. Repeat, forever, across thousands of requests a week.
Serval, a San Francisco software company founded in 2024, exists because its two founders stared at that fact long enough to name it. Jake Stauch, the CEO, spent five years at the security-hardware company Verkada doing customer discovery with enterprise IT departments. Alex McLeod, the CTO, was there too. What Stauch kept noticing was a paradox: the reason IT teams automated so little wasn't that they lacked tools. It was that building a workflow in those tools took more effort than just doing the task by hand. So they did the task by hand. Rationally. Thousands of times.
Serval's product is an answer to that specific piece of economics. It is an AI-native IT service management platform - ITSM, in the acronym-heavy language of the category - and its central trick is to invert who does the building. Instead of an admin painstakingly wiring up a workflow, the admin describes what they want in ordinary English. Serval's engine turns that description into executable code. "Offboard this employee" becomes a workflow that deactivates the user in Okta, logs them out of every Google Workspace session, revokes third-party app access, and files the record - each step visible, each step something a human can read before it runs.
That last part is the quietly important part. Serval is not selling a black box that promises to figure it out. It generates workflows you can inspect, version, and audit. In consumer software, "just trust the model" is a viable pitch. In enterprise IT - where the buyer often has a compliance team standing behind them - "here is exactly what it will do, in code" is a much better one. Serval leans into that with the unromantic machinery enterprises actually ask about: SAML, SCIM, role-based access control, audit logs, encryption in transit and at rest, and the option to self-host rather than send everything to someone else's cloud.
The results customers report are the kind of numbers that make investors sit up. Companies using Serval say they are automating more than half of all their IT tickets - not just password resets, but access provisioning, onboarding, offboarding, and reporting. Some have gone further and replaced their previous ITSM system entirely, making Serval the system of record for how operational work gets done. That is the difference between being a feature bolted onto a legacy platform and being the platform.
Which brings us to the money, because in 2025 the money arrived in a hurry. In October, Redpoint Ventures led a $47 million Series A to, in the category's phrasing, "bring AI agents to IT service management." Roughly two months later - before most companies would have finished spending the first check - Sequoia led another round of $75 million at a reported valuation of about $1 billion. Total raised now sits in the neighborhood of $122 to $127 million, depending on which tally you trust, gathered in well under two years.
Preemptive rounds like the second one tend to happen for a reason, and Sequoia was unusually candid about it. The firm said it moved to get ahead of the financing after hearing customer feedback so strong that partner Anas Biad reached for a comparison from 2009: the last time Sequoia heard signal this loud, he said, was when it backed ServiceNow. That is a large thing to say out loud, because ServiceNow is the incumbent Serval is implicitly trying to unseat - a company now worth many tens of billions of dollars, built on exactly the workflows Serval wants to rewrite in English.
None of this is guaranteed to work, and it is worth being honest about that. The AI-for-IT space is crowded and getting more so. Moveworks was recently acquired by ServiceNow, which means the biggest incumbent is now actively fusing agentic AI into its own platform. Aisera advertises auto-resolution rates of its own. Atomicwork, Espressive, and Redwood are all pursuing overlapping slices of the same problem. Serval's differentiation - natural language that compiles to auditable code, deployable on your own infrastructure - is real, but "real differentiation" and "durable moat" are not the same sentence, and the category will spend the next few years arguing about which one Serval has.
What is notable is how far Serval has already pushed past its starting point. The company began with IT, but the pattern - describe a repetitive back-office process, get an auditable automation - turned out to generalize. Teams in HR, Finance, Legal, Security, and Engineering began asking for the same thing, and Serval expanded to serve them. Over the stretch around its two big raises, the company reports growing revenue by roughly 500% and more than tripling its headcount, which now sits around 130 people. Referenced customers include the AI search company Perplexity and, fittingly, Verkada - the place where the founders first watched the problem up close.
There is a temptation, when a company raises this much this fast, to narrate it as inevitable. It wasn't. Stauch's previous company, NeuroPlus, made brain-sensing hardware and cognitive-performance software; he ran it from 2012 to 2019 and landed on the Forbes 30 Under 30 list in 2017. Hardware and consumer neuro-gadgets are about as far from enterprise IT ticketing as a founder can travel. The throughline is less the domain than the temperament: Stauch has talked, on more than one podcast, about a "go hard early" philosophy and about betting on a technology before it fully works, then building the boring infrastructure - reliability, guardrails, auditability - that makes it safe to hand real authority to an AI agent inside a real company.
It is worth pausing on why "auditable code" is the whole ballgame and not a footnote. An AI agent that resets a password is a convenience. An AI agent that can deactivate accounts, revoke access, and change who can see what inside a company is holding a loaded set of permissions, and the person who signs off on giving it those permissions is not an engineer excited about AI - it is a security lead who gets fired if something goes wrong. That person does not want a probability distribution. They want a diff. Serval's architecture, where a plain-English request compiles down to versioned workflow code with multi-step approvals and role segregation, is really a way of speaking that person's language: not "the model will probably handle it," but "here is the exact sequence, approve it or don't." Multi-step approval workflows and audit logs are not features bolted on for a compliance checkbox; they are the reason the software is allowed near the systems that matter.
The name, for what it's worth, is doing a small amount of work too. A serval is a mid-sized African wild cat known for an absurd vertical leap - it hunts by springing above tall grass and coming down on prey before the prey knows it is being hunted. As corporate metaphors go it is more apt than most: the pitch is to catch the routine request in the air, before it lands in a queue and waits a day for a human. Whether that is branding genius or a coincidence the founders backfilled, it captures the actual product behavior, which is more than most startup names manage.
So what should a reasonable person conclude about Serval in the middle of 2026? That it has found a genuinely sharp wedge into an enormous, unglamorous market; that it has convinced serious customers to make it their system of record rather than a toy; and that it has raised enough money, fast enough, to be taken seriously by the incumbent it is aiming at. Also that the hard part is still ahead. Turning a fast start into a durable business - through a crowded field, against a ServiceNow that is now buying AI companies of its own, while keeping the auditability promise intact as customers hand agents more authority - is a multi-year problem, and no amount of Series-something enthusiasm skips it. Serval has answered the founding question convincingly. The rest of the questions get asked one automated ticket at a time.
That is ultimately the wager Serval is making, and it is a more interesting one than "chatbot for tickets." The bet is that the place enterprise AI first pays for itself is not the flashy demo but the invisible work: the offboarding nobody wants to run, the access request that sits in a queue for a day, the report someone rebuilds every Monday. It is unglamorous, it happens thousands of times a week, and - if Serval is right about the economics - it is worth about a billion dollars to make it disappear. Whether that number holds is a question for the next few years. The question Serval has already answered is a smaller and more useful one: why does a company full of automation tools still do everything by hand? Because building the automation cost more than the labor. Change that, and a lot of hands are suddenly free.
