The $1B AI Agent That Smells Like 2017 — and Why It's a Bigger Threat to ServiceNow Than to Crypto

Over the past week, I watched a valuation bubble form — not in crypto, but in enterprise AI glue. Serval Systems, a San Francisco startup you've never heard of, just raised a Series B at a $1 billion valuation, led by Sequoia. Total raised: $127 million. The product? Catalyst, an AI agent that reads your IT ticketing history, generates TypeScript workflows, deploys background agents to monitor your systems, and proposes fixes before humans notice anything broke. Sound familiar? It's the nightmare echo of 2017 — except this time, the token is a workflow, and the ledger is... well, nothing. Yet. The boldest claim in the press release: customers using ServiceNow's AI products have an actual deployment rate under 10%. ServiceNow denies it. But in the hush before the storm, that's the most important number. Let's rewind. ServiceNow is the dominant player in IT Service Management — the digital brain that enterprises use to track tickets, manage incidents, and route requests to humans. For two decades, its playbook has been "platform lock-in": charge for modules, bolt on a low-code workflow builder, and then charge again for implementation specialists to make it all run. It's the Wall Street of IT operations — centralized, heavy, and resistant to change. Catalyst attacks that church from the side door. Instead of dragging boxes on a canvas, you let an AI generate the entire workflow from raw ticket data. TypeScript, not drag-and-drop. Code, not config. That's not an incremental upgrade; it's a paradigm flip from "human defines rules" to "AI proposes rules, human approves." If you've been in crypto for more than a year, this pattern should set off every alarm you have. In 2017, I spent 72 hours scraping 0x Protocol's relayer network, looking for liquidity shifts before the crowd noticed. I learned that the real alpha hides in the gaps between data sources — the order flow that didn't match a headline, the address that moved just before a listing. Catalyst's alpha hides in the same way: the unloved ticketing history, the Slack threads, the system logs that were never normalized. The press release tells you that Catalyst analyzes ticket history, identifies recurring patterns, drafts workflows, skills, forms, access policies, and dashboards, and then creates background agents that continuously watch connected IT systems. That's a full pipeline: discovery → generation → execution. But here's what the glossy announcement doesn't say: the model layer is rented. Serval doesn't train its own frontier models; it calls someone else's inference APIs. The real moat isn't the GPU cluster. It's the private ticket corpus and the system integration layer. That's a data moat, not a model moat — and in my seven years watching decentralized systems, data moats can evaporate when a competitor plugs into the same ticketing APIs. The customer cases are solid but narrow. Ramp, a fintech company, says it builds workflows 50% faster with Catalyst and has expanded from IT to about ten teams, including finance, legal, and business operations. That's real product-market fit — but Ramp is a tech-savvy customer, the kind that would have built its own internal tools a decade ago. Mercor is an outsourcing platform; that's a long way from a Fortune 500 factory floor. The headline number — $1 billion valuation — implies an ARR somewhere between $10 million and $30 million if you use standard 20-40x multiples for AI-native SaaS. At $20 million ARR, that's a 50x price-to-sales ratio. In traditional SaaS, that's insanity. In the current AI hype cycle, it's "normal-ish." But it means Serval needs to hit $50-100 million ARR in the next 18-24 months just to justify the round. Speed is the currency, but accuracy is the vault. Now the part everyone in crypto should care about: Catalyst's core mechanism mirrors a smart contract — except it's missing the ledger. When an AI agent writes TypeScript to automate a fix, it's executing state changes on infrastructure. That's exactly what a smart contract does on a blockchain, minus the transparency. In crypto, we solve the trust problem with immutability, audit trails, and deterministic execution. Catalyst solves it with a human review checkpoint: all generated artifacts go to a human first, then get published. That's a Human-in-the-Loop safety net, and it's better than nothing. But once trust builds, humans rubber-stamp. We've seen this pattern in every automated compliance system. The moment "AI proposes, human reviews" becomes "AI proposes, human nods," you've reintroduced the bug class we've spent years trying to kill: unauditable state changes. What happens when a background agent misreads a warning as a critical failure and restarts a production database? Who's liable — the vendor, the customer, or the model nobody can interrogate? The industry has no answer yet. That's the blind spot. And it gets more uncomfortable. Catalyst's agents "continuously check connected IT systems" and can generate access policies. That means an AI is being handed the keys to identity and permission management. If a malicious actor compromises the agent's connector, they're not just reading tickets — they can laterally move through privileged APIs, escalate access, and trigger automated workflows to cover their tracks. We spent 2021-2024 arguing about Dao hacks and bridge exploits. The enterprise version of that catastrophe is an AI-driven supply-chain attack on an ITSM platform that controls hundreds of downstream systems. The damage would make the Ronin bridge look like pocket change. The mainstream analysts are framing this as Serval vs. ServiceNow. That's the surface narrative. The contrarian angle is that the real fight is against a four-front war: Microsoft Copilot Studio, UiPath's agent pivot, Atlassian Intelligence, and the consulting giants (Accenture, Deloitte) whose implementation businesses are threatened by any tool that generates workflows automatically. ServiceNow is the visible enemy, but the silent ones are the ones with distribution. Microsoft can bundle an AI workflow generator into Microsoft 365 and price it at zero marginal cost. That's a knife that doesn't need sharpening; it just needs mass. And don't forget the weirdest thread: ServiceNow acquired Moveworks in December 2025 for $2.85 billion. Moveworks customers who don't want to be folded into the mothership are an orphaned pool of AI-native IT buyers — exactly Serval's target. Nobody's talking about that. Alpha leaks in silence. The deeper truth is that AI-generated workflows are shares of company infrastructure — but they're not auditable in any decentralized way. Ask Serval: can you prove that a specific workflow change came from a specific model version, at a specific timestamp, with a hash of the input data? No. Because there's no ledger. That's where blockchain actually enters this story — not as a payments rail, but as a provenance layer. If every workflow artifact was anchored to a blockchain, you'd have a tamper-evident audit trail for every automated change. That would solve the accountability problem that currently keeps Serval from selling into heavily regulated industries. Ironically, the company that will benefit most from distributed ledger technology might be Serval's own compliance team. Echoes of 2017 whisper through every new bull run. What matters next is the migration path. In the next 12-24 months, the enterprise IT market will split into two lanes: large enterprises that stay with ServiceNow because migration cost is too high, and mid-sized forward-leaning companies that adopt AI-native alternatives because they're cheaper and faster. That split resembles the old Snowflake-versus-Teradata battle — it took a decade to play out. Serval has about 18 months of runway before it needs a Series C, based on a typical $50-80 million annual burn. If the market mood turns cold, that $1 billion valuation could be a ceiling, not a floor. But if Catalyst's agent reliability improves and it gets SOC 2 Type II and FedRAMP certifications, the acquisition premium becomes real. A 15-25x revenue acquisition from Microsoft or ServiceNow is not a pipe dream if the tech proves itself in production. The lesson from my audit day job: when I see a protocol with 90% of its deposits in one liquidity pool, I short it. When I see a startup with a 50x price-to-sales ratio and a single marquee customer, I watch it with suspicion — but I also recognize the pattern. The 2017 ICO mania gave us Uniswap. The 2026 enterprise AI mania might give us the first truly autonomous IT infrastructure. The winners will be those who treat generated code like unaudited smart contracts: paranoid, reviewable, and reversible. If they don't, the market will eat them alive. And if they do, the next big bull run won't be on a chain — it'll be in the server room. The watchpoint is the deployment-rate war. Every quarter, someone will release a stat about how little AI shelfware is actually used. That's not a technical metric; it's a narrative weapon. Don't blink. The ledger doesn't forget.