Her AI did 20 hours of work in a 10-hour day. That was the problem

Joni Pirovich founded APAC's leading blockchain and digital assets specialist law firm. She has acted for emerging tech platforms and for foreign governments. And for years she ran it as a solo principal, servicing an industry where the law changes faster than any one lawyer can read it.

Her answer was to build. Crystal Chat was a generative AI tool that spoke to her own library of global crypto-specific laws, cases and regulatory guidance — the thing she had been holding in her head and her files.

It did what she built it to do. That is the part of the story most founders would stop at.

Roughly six months in, the maths stopped being encouraging and started being a warning. The tool was compressing about twenty hours of work into an eight-to-ten-hour day — which meant the ceiling was no longer the work. It was her.

A human driving a generative AI tool, she decided, was not a sustainable design. It was a faster version of the same bottleneck.

So she went back to the drawing board — not to improve the tool, but to work out what a system would need to contain before it could run without her hand on it.

What came out of that is the Crystal Agentic Operating System, built around what she calls Agentic Assets: containers for the knowledge, skills, experience and processes a piece of regulated work actually requires. Her analogy is a self-driving car — autonomy that has to be earned through oversight, evaluation and continuous improvement rather than declared on day one.

This is the argument the whole company rests on, and it is aimed squarely at how easy building has become.

“It's very easy and fun to use AI to build apps, to do tasks, and to do parts of your job. But what most people don't realise is that the ongoing effort to maintain your apps and keep them secure and working requires so much more under the hood.”

For a law firm or a financial services firm, the bottom of that iceberg is not a technical detail. It is the regulator's question.

Crystal aOS is sold as the outcome first and the capability second. A firm can start by buying the function, month to month, and only take it in-house once it knows what it is taking on.

“At Crystal aOS, we provide a compliance back office for regulated firms that need their system and operations to be safe and secure.”

When the firm is ready, her team goes in and installs the AI-native compliance function — and, critically, teaches the human team the maintenance work that keeps it safe. The framework itself is built to align with ISO 27001 and ISO 42001, and comes with the policies, SOPs, templates and registers a governed AI operation needs.

Crystal aOS is not a product she sells and then works around. It is the way her own business operates.

“I use AI from wall to wall in my business today. It helps me to plan my day, to understand the market, to understand my competitors, but also to deliver my services and keep my knowledge artifacts current.”

She is still the one signing off. What has changed is that she is no longer the first responder to everything that arrives.

“We've been working with a few early clients now, and we've battle tested the system in our own operations and with those in regulated industries. So we're confident that the model works, and we're now testing that the model can scale.”

Through the AI Pathfinder program she describes getting access she would otherwise have had to assemble herself — developer and coding days, founder strategy days, and the practical things a solo founder ends up doing alone.