AI in Legal Practice: Gunjan Paharia on the Future of Lawyers and Law Firms
AI is steadily becoming part of everyday legal practice, changing how lawyers research, draft, manage information and serve clients. In this written interview, we asked Gunjan Paharia, Founder & Managing Partner of ZeusIP Advocates LLP, about how AI is reshaping legal practice, the risks firms need to consider, and what the future could look like for lawyers and law firms.
AI is increasingly moving from an experimental tool to part of day-to-day legal practice. What do you see as the biggest opportunities and the biggest risks for law firms as they integrate AI into legal workflows?
We’re still at an early stage. The biggest opportunity is time — AI frees lawyers from volume work and structures it, so we can get closer to spending more hours on strategic and critical thinking (and, frankly, getting home on time). The biggest risk is output quality. LLMs tend to produce generic boilerplates, and cleaning that up can eat more time than it saves. In my experience, AI performs well only when it’s working from well-built artefacts and templates designed for it to follow. The other major risk is hallucination — in law, an unvalidated hallucination can be the death knell for a client’s case. My firm belief is that lawyers should build and train their own agents rather than use off-the-shelf LLMs blindly — which is exactly why we built RIA in-house on our own infrastructure rather than licensing an off-the-shelf legal AI product.
As AI takes over more research-intensive and repetitive tasks, how do you see the role of lawyers — particularly young and junior lawyers — changing? Could over-reliance on AI also affect how the next generation develops legal skills?
Absolutely! I The way lawyers work has to change dramatically, and the hardest part is the law school curriculum. Schools worldwide are slow to adapt, and that’s a real problem. My concern is that students are still being trained for the practice as it existed in the early 2020s, not for how it will look in 2030. Until law schools catch up, the responsibility falls on students themselves — to track developments in technology in real time, upskill continuously, and learn to think laterally.
Client confidentiality and data security remain major concerns around AI adoption in law. What safeguards should law firms have in place before allowing AI tools to work with sensitive legal documents and client information?
The necessary guardrails already exist in today’s technology. Using generic, boilerplate LLMs for document or evidence review isn’t a safe practice. The way forward, in my view, is to keep AI — or a dedicated version of it — within the firm’s own environment, operating in a tightly controlled loop. In practice, that’s exactly how RIA is built: its core system is trained entirely on ZeusIP’s own data and runs on infrastructure we control, so it is not learning from anyone else’s information and client data does not leave our systems for that part of the work. Certain narrower tasks — open-source research on a party, for instance — do draw on external tools, the way a lawyer would otherwise do that research manually; we have opted out of model training wherever providers allow it, and we are fully compliant with the DPDP Act and GDPR, including client consent wherever data may be processed on servers outside India.
Where should the line be drawn between AI-assisted legal work and decisions that must remain firmly within the domain of professional legal judgment? Who should ultimately be accountable when AI-assisted work goes wrong?
The lawyer is always accountable. They own the brief and the advice they sign off on. AI is an enabler, not the doer, and that distinction matters. I often hear people say “AI did it” — the narrative our industry needs to embrace is “AI helped me do it.” That should always hold true for law.
We’ve actually built that principle into a formal internal protocol we call F.I.V.E. — Fact-check, Improve, Validate, Enterprise. The first three steps are quality control: checking outputs against source documents, tightening language, validating against our own practice standards. Enterprise is the step I find most interesting — it’s where the lawyer steps back and thinks about the client’s real objective, not just what the document says. Nothing reaches a client or a filing without that sign-off.
With the time AI frees up, lawyers should invest in thinking, observing, mastering courtcraft, finding creative approaches to drafting, sharpening advocacy, and deepening their understanding of jurisprudence. There’s a lot to do — and now, finally, the time to do it.
What gaps in existing IP research and litigation workflows did ZeusIP identify that prompted the development of RIA, particularly across trademark, patent and contentious matters?
There wasn’t really a gap so much as a bottleneck. Lawyers spend significant time on data and information gathering — time that is a finite resource and better spent on real advocacy and skill-building. If there was a genuine gap, it was institutional memory. It’s humanly difficult to recall documents, evidence, and mandates across client portfolios that can run into hundreds, even thousands, of files. We’re talking about roughly eight million documents built up over more than two decades of practice — a lot of that experience simply wasn’t accessible day-to-day. A client would ask for a chances-of-success view and we’d have to manually reconstruct context a senior partner already knew intuitively. Clients were being asked for evidence they’d already given us, sometimes years earlier, because finding it in an old file took longer than just asking again. And an associate joining today had no way to know we had made a particular call on a similar conflict five years ago, unless the partner who remembered it happened to still be around. RIA solves that — it remembers, searches within minutes, and retains full context across files.
As IP matters increasingly involve large volumes of case law, filings and other documents, how is the way lawyers approach research and case preparation changing? What challenges arise when practitioners need to locate and assess information across such large repositories?
Have already answered this above.
Do you see AI eventually moving beyond being a productivity tool to becoming something closer to a “digital associate,” capable of handling increasingly complex layers of legal research and analysis? Where would you still draw the boundary?
I would say it is already there. RIA isn’t just a tool — it functions like a persona, a digital associate. It reads files, understands context, drafts documents, assesses chances of success, and advises. On something like a chances-of-success question, for example, RIA can lay out twenty-plus data points — what’s in the client’s favour, what isn’t — but it doesn’t assign that a percentage, and it doesn’t decide whether we file. The boundary is that it is not permitted to act autonomously. Its workflows are deliberately structured so it cannot reach an outcome without manual intervention at key checkpoints, and the final output is always fact-checked, refined, and validated by a human who has the final say — that’s the Enterprise step of our F.I.V.E. framework in action.
AI is also creating new legal questions around ownership, inventorship, copyright and protection of AI-generated works and inventions. Which of these issues do you believe will become the most consequential for IP lawyers and businesses over the next few years?
The slow pace of regulation is a challenge, but jurisprudence — as with most areas of law — develops through the courts, grounded in real cases, and that’s already underway alongside the work of think tanks and governments. Consider the “work for hire” doctrine: it holds that an employee’s work belongs to the employer because the employer directed the work and paid for it. Apply that logic to AI — a human directs the output through a prompt and pays for it, whether through tokens or a subscription. Why should not that be treated the same way, as work for hire?
Could widespread AI adoption fundamentally change the economics and delivery model of legal services — including how firms price work, structure teams and measure lawyer productivity?
That does appear to be the direction the industry is heading. These conversations are already happening, and barring social or regulatory intervention, the future will look very different from what we’ve known. Most, if not all, law firms will need to build robust in-house tech teams. There’s also a growing idea gaining traction — that tech companies may start hiring lawyers of their own.
