Dominic is the CEO of Lawpath, dedicating his days to making legal easier, faster and more accessible to businesses. Dominic is a recognised thought-leader in Australian legal disruption, and was recognised as a winner of the Australian Legal Innovation Index and recently a winner of the LexisNexis 40 Under 40 (APAC).
At Lawpath Legal, our lawyers keep our legal AI model current by designing a loop around the practice. Every advice call they finish becomes material for Cygnet, our own model, and any document generated by the firm goes back into it too. Loop engineering is the design of that loop: what starts it, how the answers are checked, and what is allowed back into the model, so the knowledge can compound. We own both the model and the practice, so the two improve each other in real time.
A law firm’s knowledge mostly lives in its lawyers. If you’re paying for advice, you’d hope the firm learnt something from every business that asked your question before you did. In most firms, that depends on which lawyer you get.
This is part 2 of a series. Part 1, What an AI-native law firm is, and how to test one, sets out what we mean by the label and four questions you can ask any firm that uses it.
- The loop starts on the call. A completed advice call is the trigger. Topics become questions, which are researched and verified before anything goes back into Cygnet.
- Firm documents go back in too. Any document generated by the firm is also fed back into the model.
- The material adds up. Lawpath transcribes hundreds of hours of legal advice call transcripts every week.
- Lawyers still set the bar. They own the 340-test suite every version of Cygnet has to pass. Lawpath Legal is a regulated incorporated legal practice with professional indemnity cover.
Where does a law firm’s knowledge usually go?
Think about the last time you rang a lawyer. Chances are they worked something out on that call that would help plenty of businesses besides yours.
In a traditional firm, that knowledge stays in one lawyer’s head. It’s hard to share across the firm.
At Lawpath Legal, our lawyers work with a model we own. What one of them learns on a call can go into that model, and from there it’s shared across all of our lawyers. Each call adds to what the next one starts with.
That’s the part a traditional firm can’t easily copy. Knowledge that builds on itself.
How does an advice call feed back into our legal AI model?
Nothing goes straight from a call into the model. After a real lawyer completes an advice call, there’s a settling period. Then the work starts:
- The topics spoken about on the call are turned into questions.
- Agents research the answers in legislation, case law and Lawpath’s other databases.
- Those questions and answers are verified and grounded in source truths we know are correct.
- Only then does the material go back into the model.
The completed call is the trigger. Agents research the questions. The verification step is the one that matters most. A model that learns from everything it hears will learn the mistakes too. That’s why nothing from a call reaches ours until it has been checked against a source we trust. Our lawyers stay in the loop by design.
Then there’s scale. Lawpath transcribes hundreds of hours of legal advice call transcripts every week. Every call generates more material for the model to learn from, so the loop compounds.
Alongside the call transcripts, any document generated by the firm is also fed back into the model.
Mistakes caught on real client matters are anonymised and worked back into the model so they don’t happen again.
One short note on privacy. Customer data is anonymised when it’s used to train and reinforce the model, and that covers the call material and the firm’s documents.
Why does owning the model and the firm matter?
AI helps our lawyers draft and review. Our lawyers, in turn, keep the model current. Each side makes the other better at its job. That’s what loop engineering looks like when you own both the model and the firm.
Picture a business owner opening Atlas, our AI system, at 11pm. Atlas is the same system in that Lawpath product and in the tools our lawyers open the next morning, and our lawyers get more advanced Atlas tools built for legal work. When your matter reaches one of them, the documents and history arrive with it.
A software business with no practising lawyers finds out what it got wrong when a customer complains. A law firm with no software ends up serving the clients who can afford an hour of someone’s time. Owning both means we can improve the model and the services in real time, and both compound.
That’s one advantage of being an AI-native law firm. Lawpath Legal also works under the same roof as our accounting and compliance firms, Lawpath Tax & Advisory and Lawpath Compliance.
What standard does the model have to meet?
Any loop that compounds needs a standard to compound against. Our lawyers set it.
They write the rubrics used to judge Atlas, and they own the 340-test suite every version of Cygnet has to pass. Our lawyers settled what a correct answer looks like before the training data for Cygnet was generated. That’s possible because we have years of history and experienced lawyers.
One rule matters more to a law firm than the rest. When a matter needs tailored advice, handing it to a lawyer or tax professional is scored as the correct answer. The model gets rewarded for knowing where its job stops.
How did Cygnet score?
On the 340-test suite, Cygnet failed 2 tests. The frontier model we had previously run in production failed 4.
The frontier model still did better on broader tasks, and the judges preferred its writing. We published that next to our results in the engineering write-up.
For a law firm, that trade is easy to make. A polished answer can still skip a required search or point to a template that doesn’t exist. The tests are built to catch exactly that.
What happens when the rules change?
Knowing a rule changed is the easy part. Knowing what it means for your business is a professional’s job.
Changes from ASIC and the ATO feed into Atlas, so a new rule can reach the system without waiting for the model to be retrained.
Payday super is a good example. From 1 July 2026, employers have to pay super at the same time as wages instead of quarterly, and the ATO’s Payday Super guidance for employers sets out the detail. A general model trained before the change may still say super is quarterly, and sound certain doing it.
Our professionals interpreted the payday super changes and fed that back into the model.
What we are not claiming
That the loop makes the model right every time. Cygnet still failed 2 of the 340 tests. One was an over-eager suggestion to book a consultation and upgrade a plan, attached to a simple GST question.
That Cygnet beats frontier models in general. It doesn’t, and we’ve published where it falls short.
That our results say anything about anyone else’s model. They’re internal tests on Lawpath scenarios.
Nor that software can stand in for a lawyer’s judgement on a matter that needs one. Atlas is built to hand those over.
Legal AI model FAQs
How does Lawpath keep its legal AI model current?
Our lawyers do it through their work. Topics from each completed advice call become questions that are researched, verified against sources we know are correct, and only then fed back into Cygnet, our own model. Any document generated by the firm is also fed back into the model.
What does loop engineering mean for a law firm?
It means the practice is designed as a loop. A completed advice call is the trigger. Agents research the questions, and nothing goes back into our legal AI model until it’s verified against a source we trust. Our lawyers stay in the loop and own the tests every version of Cygnet has to pass.
What happens after a Lawpath advice call?
There’s a settling period. Then the topics from the call are turned into questions, agents research the answers in legislation, case law and Lawpath’s other databases, and the questions and answers are verified before anything reaches the model.
Who sets the standard the model has to meet?
Our lawyers. They write the rubrics used to judge Atlas and own the 340-test suite every version of Cygnet has to pass.
Does Atlas give legal advice?
Atlas gives general information and drafting support within its scope. When you need advice tailored to your situation, it passes the matter to a qualified professional, and a Lawpath Legal lawyer makes the call.
Is Lawpath Legal a regulated law firm?
Yes. Lawpath Legal is a regulated incorporated legal practice with professional indemnity cover.
You don’t need to follow every step of the loop to use Atlas. What matters is that each call makes the next answer a little better, and a lawyer still makes the call when your matter needs advice.
Every call teaches the firm something. Now the whole firm gets to keep it.
Credits: written and reviewed by Dominic Woolrych, CEO of Lawpath.
If your business has a question that needs a lawyer’s judgement, talk to Lawpath Legal. We’ll match you with the right lawyer and quote before work begins.
