COO Vimal Choudhary on How Khaitan & Co is Putting Its Own AI to Work

At Khaitan & Co, we have been deliberate about how we approach AI, treating it as a tool that can genuinely enhance the quality of legal work, rather than something to adopt simply because it is being talked about everywhere. We built proprietary infrastructure to deploy AI in a controlled, secure environment before rolling anything out more broadly, which gave us the space to learn without compromising client confidentiality or compliance standards.

Our AI platform, KAI, has now matured into KAI 2.0, and what makes it meaningful is that it was built around how our people actually work. It supports initial research, document review tasks, and also helps members develop precise prompts and purpose-built AI agents for their specific team needs. But the platform is only part of the story. The bigger investment has been the training and sensitization of people, ensuring that members, regardless of their comfort level with technology, understand how to use these tools well. When AI stops being something we experiment with and becomes part of how we naturally work each day, that is when the real impact will start to fully show.

The areas with the clearest near-term potential are initial document review, legal research, and automating specific time-intensive internal processes. We have already started deploying tools in some of these areas, though we have been intentional about not rushing the rollout. Getting the training right and testing it before scaling matters more to us than being first.

For research and document review, I expect meaningful scale within the next few years. For more nuanced legal work, the kind that requires deep contextual judgment, professional expertise, and knowledge of how law actually operates on the ground, the timeline will be longer, and rightly so. The goal is not to automate legal work; it is to free up our lawyers to focus on the parts of their work where their expertise is irreplaceable.

KAI 2.0 is an internal platform. It does not interact with external servers, and sensitive client information stays within our own infrastructure. In that sense, yes, it functions as a secure sandbox where members can explore, experiment, and build familiarity with AI without any risk to data security or client confidentiality. Our compliance protocols are built into the platform itself, not layered on top of it as an afterthought. This architecture was a conscious decision; we wanted a foundation that members could trust before asking them to embed it into their work.

The pace of change is faster than most people in the profession are willing to publicly acknowledge. Within the next two to three years, I expect AI to be doing substantive first-pass work on contract review, due diligence, regulatory mapping, and litigation research, not as an experiment, but as a standard part of how legal work is done. That is not a distant horizon; some of this is already happening.

What excites me more is what comes next. AI that can synthesize large bodies of precedent to surface risk patterns a lawyer might take days to identify, tools that can model how a regulatory change affects a client’s entire contract portfolio simultaneously, or systems that flag emerging compliance exposure before the client even knows to ask the question. These are not incremental efficiency gains; they change the very nature of the legal advice.

At the same time, none of this displaces the lawyer’s judgment. It sharpens it. The lawyer who uses these capabilities well will consistently outperform the one who does not, and that gap will widen quickly. The firms that will define the next decade are the ones investing now into building those capabilities thoughtfully and at scale.

Across the legal profession broadly, the challenges are fairly consistent regardless of firm size or geography. The technology itself is often the most straightforward part. The harder work is cultural and organizational.

The first challenge is trust. Lawyers are trained to be precise and accountable, and AI hallucinations, where a system produces confidently wrong output, cut directly against that instinct. That concern is legitimate and should not be dismissed. The way to address it is not to oversell AI’s reliability, but to design workflows where human oversight is genuinely built in, and to be transparent with practitioners about where AI tools are strong and where they are not.

The second challenge is relevance. Generic AI training programs that do not connect to how a specific practice group actually works tend to produce polite attendance and limited adoption. What drives real change is when a lawyer sees a tool solve a problem they personally experience every day. That is the moment skepticism turns into advocacy.

Hallucinations, that is, AI generating plausible-sounding but factually incorrect information, top the list, and with good reason. In legal work, the cost of an undetected error is high. The answer is not to pretend this limitation does not exist, but to design workflows where human review is built in as a non-negotiable step, not as an optional check. Right now, all AI-assisted work at the firm involves careful human oversight, and we expect that to remain the case for the foreseeable future.

The broader concern underneath the hallucination question is really about trust. Lawyers want to know that a tool will not quietly make their work worse while appearing to make it faster. Building that trust takes demonstrated reliability over time, transparent communication about what AI can and cannot do, and an organizational culture where raising concerns about a tool’s output is encouraged rather than seen as being resistant to change.