I (Céleste Urech) started in law and left pretty quickly. At the time I told myself it was because everything moved too slowly. Decisions took forever, nobody seemed interested in doing things differently, and I wanted to build things. So I went and did that — staffing, media, digital products. Learned to ship software, run teams, figure things out on the fly.
It took years before I realized I’d been unfair to law. The slowness isn’t a bug. Law underpins trust, property, markets. Of course it resists unproven ideas. The cost of getting it wrong is too high to move fast and break things.
But there’s a difference between careful and just… inefficient. That’s where CASUS came from.
Coming back
My co-founder Fabian Staub and I kept coming back to the same nagging question: why are legal professionals, some of the most precise thinkers you’ll meet, still spending so much time on repetitive document work? The tools available to them were either too generic, too insecure, or clearly built by people who’d never sat through a contract review.
We started building infrastructure for legal teams. Not an AI demo, not a “replace lawyers” pitch.
Getting firms to try it was harder than building it
The objections we heard were legitimate. Data protection. Attorney-client privilege. Hallucinations. Regulatory gray areas. And a very reasonable skepticism toward tech founders showing up to “fix” law.
Honestly, that skepticism was deserved. These are people responsible for their clients’ trust. I’d be suspicious too.
Then LLMs changed the conversation
Large language models shifted things, because they made it obvious that language can be processed structurally, and law is mostly structured language. The cautious move stopped being “wait and see.” Waiting became the risk.
I started hearing different questions from lawyers. Less “why would we ever use this?” and more “how should we evaluate it?” and “what governance do we need around it?” Firms started forming AI task forces. In-house teams ran pilots. Bar associations began drafting guidance. Some lawyers even started vibecoding the tools they needed.
The mood changed.
What CASUS actually is
We’re not a chatbot wrapper. We build one thing and we build it well.
AI-powered contract review and drafting: teams can check contracts against their own benchmarks, extract structured data, generate summaries, draft clauses based on their own knowledge. The principle we won’t compromise on is that AI assists and lawyers decide. Everything is transparent and reviewable. No black box.
On security
In legal tech, if security isn’t right, the rest doesn’t matter. Swiss and EU hosting, strict data separation, no data retention for AI processing, no training on customer documents. We treat that as table stakes.
Building with lawyers
I’ve learned this the hard way: if you build legal tools without lawyers deeply involved, you’ll build the wrong thing. The workflows are full of edge cases that seem irrational until someone explains the regulatory logic behind them.
So we test with real contracts, refine with real clauses, and iterate based on how associates actually work, which is often late, tired, and on their fourth document. If the tool doesn’t help in that state, it doesn’t help.
Why I’m building this
I left law because it felt too slow. I came back because I understood why it has to be. I’m building CASUS because the technology is finally good enough and the industry is genuinely ready to engage with it.
Legal teams that figure this out will spend more of their time on judgment and less on formatting. That’s the layer we’re trying to build.
