Advocate, a New York risk management firm, had a problem: 65,000 mortgage insurance policies stacked up, unprocessed. Manual review would have taken a decade. The backlog disappeared in three weeks.
The system behind that clearance, Felix, launched publicly on Tuesday with £1.7 million in backing and a pitch that challenges how law firms are actually using artificial intelligence.
Tomas Scavnicky built the platform after selling his previous company, Parrot, to legal software provider FileVine in 2025. Where Parrot focused on workflow automation for law firms, Felix tackles what he calls “shadow AI”—the unmonitored, prompt-based usage spreading across professional services without governance or consistency.
“AI has made it easier to work faster as individuals, but it hasn’t solved how work gets done across a business,” Scavnicky said. “In professional services, you need consistency, accountability and control. Felix turns AI from something experimental into something you can actually run your operations on. In law, that matters. You’re not relying on black box outputs, and your clients get both the service and the confidence they expect.”
Around 70% of firms now use AI in some capacity, according to industry surveys. Yet much of that adoption happens outside formal systems—lawyers firing prompts into ChatGPT or Claude, getting different answers each time, with no audit trail and no way to guarantee the same input produces the same output twice.
Felix takes the opposite approach. Users describe processes in plain language. The platform converts those descriptions into structured workflows that run identically every time, logging every step and decision. Same input, same output. Always.
That determinism matters in sectors where decisions carry legal, financial or regulatory weight. A probabilistic answer from a foundation model won’t cut it when compliance officers ask how a conclusion was reached six months later.
The company raised its £1.7 million pre-seed round from XYZ Venture Capital, with angels including current and former executives from Amazon, Apple, Palantir, Flexport, Yelp and Midjourney participating. Several major legal, finance and insurance firms are already running the platform in production.
Scavnicky’s co-founder, Matej Vetrak, serves as chief technology officer. “Most tools today help individuals move faster. What’s missing is coordination,” Vetrak explained. “Felix allows teams to capture their expertise and turn it into systems that run consistently, with humans involved where it matters. That’s how you scale quality in environments like law.”
The platform generates and executes code automatically, integrating multiple AI models where needed whilst maintaining what the company describes as a “full audit trail” of inputs, reasoning and outputs. Firms can embed institutional knowledge—normally scattered across people, documents and legacy systems—into workflows that run whether specific employees are available or not.
For Advocate, that meant codifying the process for reviewing mortgage insurance policies into a system that operates continuously. The initial backlog vanished. More importantly, new policies now get processed as they arrive, preventing future pile-ups.
The approach reflects a broader tension in legal technology. Firms want the speed AI promises, but they also need the control and traceability that traditional workflows provide. Felix positions itself as the bridge—AI reasoning applied within guardrails, with human checkpoints inserted wherever judgement matters.
Following the FileVine acquisition, Scavnicky’s team recognised the challenge extended beyond legal into insurance, accounting and other professional services. The pattern repeated: knowledge work that’s complex, documentation-heavy and difficult to standardise, tackled by individuals using tools that don’t coordinate.
The launch coincides with a rebrand from Script.so, signalling the shift from a vertical automation tool into a horizontal platform for building AI-powered workflows across industries. The new name and positioning suggest ambitions beyond legal, though the company’s early traction remains concentrated in law firms and insurance providers.
What the funding actually buys remains somewhat vague—the company says it will “expand product capability and scale growth,” the kind of language that could mean hiring engineers or could mean aggressive sales expansion. With major firms already deployed as design partners, the commercial model appears to be working, though specific customer counts weren’t disclosed.
The platform went live Tuesday at felix.so, available to firms willing to rethink how they structure work. Whether legal departments will abandon the flexibility of prompt-based AI for the rigidity of predetermined workflows depends on how much they value consistency over improvisation.
For now, the Advocate case study provides the most concrete evidence. Ten years of work, compressed into 21 days. That’s the pitch, quantified.
Scavnicky’s track record—building Parrot to acquisition, then immediately spinning up a new company addressing adjacent problems—suggests he understands the legal technology landscape intimately. FileVine clearly saw value in his previous work. The question is whether law firms see enough risk in “shadow AI” to justify replacing ad hoc tools with governed systems.
The investor roster suggests confidence in that thesis. Angels from Palantir and Apple don’t typically back legal technology unless they see enterprise-scale potential. XYZ Venture Capital’s involvement adds weight, though the pre-seed round size—£1.7 million—indicates early-stage belief rather than market validation.
By targeting legal, insurance and accounting simultaneously, Felix avoids over-reliance on any single sector. The workflows differ, but the underlying challenge remains constant: turning expert knowledge into repeatable systems that don’t depend on individual expertise.
Whether that vision resonates widely enough to justify venture-scale returns won’t be clear until deployment numbers emerge. For now, one New York firm cleared a decade’s backlog in three weeks. That’s a start.
