For years, Indian law firms and in-house teams have heard much the same promises from legal tech: faster research, automated drafting, a lighter administrative load. Most experiments have followed a familiar arc. Initial excitement gives way to a limited pilot, and then to a quiet return to familiar ways of working.
That story is beginning to change with the arrival of India-native AI tools – platforms trained on Indian statutes, case law, circulars, and drafting styles, and deployed inside existing workflows in Word, Outlook, and firm knowledge systems. Adoption is no longer the main hurdle. The harder question now is: how do we use this new wave of AI safely, without outsourcing judgment?
In a recent Oxford essay, American philosopher Ruth Chang argues that putting the human "in the loop" of machine processing only works if the human is placed there to handle what she calls the "hard choices". These are decisions where one option is better in some respects and the other in others, and where the trade-off cannot be settled by a single metric. Oversight must be real, not symbolic.
These concerns map directly onto legal practice. Much of the work involves trade-offs that cannot be reduced to a single metric. Risk has to be weighed against reward, strategy against settlement, speed against thoroughness. Against that backdrop, human-in-the-loop (HITL) AI has gained currency among Indian firms and general counsel. The thinking is straightforward. AI is treated as an assistant that accelerates research and drafting, while lawyers stay firmly responsible for what goes out to clients, counterparties, regulators, and courts.
For a long time, the question lawyers asked of technology was a narrow one: what can we automate? In a billing-driven, deadline-heavy environment, that question makes sense. But full automation runs into practical and ethical limits quickly, and this is especially true in Indian law, where statutes, delegated legislation, and case law all shift around each other.
A human-in-the-loop model reframes the question. Instead of asking what to automate, lawyers ask where AI can assist and where a human must decide. In practice, Indian firms experimenting seriously with these tools are converging on three loops that sit around every AI interaction. There is the input, the review, and the sign-off.
The first loop is about deciding whether to use AI for a task at all, and if so, what to share and how to ask. Two issues dominate: confidentiality and relevance.
On confidentiality, most organisations now draw a hard line between consumer-facing tools and enterprise deployments. Sensitive client information remains within the firm's own environment or a tightly controlled setup, with clear commitments regarding how data is stored and used. Platforms such as August, now active across many Indian firms, are being adopted precisely because they offer these deployment models, rather than routing prompts through public interfaces.
On relevance, lawyers are learning that the quality of any output tracks the quality of the question. Vague prompts (“What is the law on arbitration?”) tend to produce generic or out-of-context answers. Specific questions (“Under Indian law, how have courts treated unilateral appointment clauses in the last five years?”) are more likely to return useful starting points.
The second loop is where most of the work sits. Here, AI output is treated as a draft, not a decision. A basic review checklist typically covers three points:
Are the citations to Indian primary materials (statutes, rules, circulars, orders, case law) clearly stated, and do they hold up against official sources?
Does the reasoning align with the lawyer's own view, the client's risk appetite, and the firm's positions on unsettled areas?
Has the output been adapted to the client's facts and industry, rather than lifted wholesale from a generic template?
Chang's point translates into a working rule for review. Human oversight is only meaningful when reviewers have both the information to spot a problem and the authority to act on it. AI tools that can surface citations by default, keep a history of prompts and answers, and allow side-by-side comparison with source documents make that oversight easier. August, for example, is built around citation-linked answers grounded in Indian law, expecting the reviewing lawyer to click through, verify, and refine before anything leaves the building.
The final loop is about responsibility. In busy practices, polished AI-assisted drafts can easily be mistaken for finished work. Firms that have written down human-in-the-loop principles are therefore insisting on a clear sign-off step. A named lawyer, usually a senior associate or partner, takes responsibility for the final version of any AI-assisted document, email, or note that goes out.
This is where the "who decided?" question gets a concrete answer. Models can do more of the heavy lifting, but endorsing a position remains human work. Some organisations have built this into their workflows directly. Where AI tools, including August, sit inside Word and Outlook, the same approval flows that already exist for important communications apply automatically to AI-assisted drafts.
As usage grows, many Indian firms and legal departments are moving from informal pilots to written AI policies. These usually identify approved tools, permitted and prohibited use-cases, data-handling rules (including DPDP compliance), and basic governance. Seen through the HITL lens, these are not compliance documents. They are institutional answers to a question every firm has to settle for itself. Where does machine assistance stop, and where does human responsibility resume?
Juniors are a particular focus. Partners worry that easy access to AI-generated drafts could short-circuit learning. The response in many teams has been to make oversight explicit. Juniors can use tools like August to generate structures and summaries. They are then expected to verify citations, adapt the language, and explain in review meetings what they accepted, what they changed, and why. Platforms that keep a clear history of prompts and outputs make these conversations concrete and verifiable.
For Indian practice, the adoption story on AI is no longer only about whether the right tools exist. They do, and they are increasingly being used on real matters by Indian firms of every size. The more interesting question is how the profession chooses to integrate them. A serious human-in-the-loop approach, informed by emerging work on oversight and grounded in everyday practice, offers a way to embrace these tools while keeping lawyers squarely in charge.
Platforms such as August now serve a growing base of Indian firms. With citation-rich answers, auditable histories, and deployment models that respect confidentiality, they show that the technology can support this model. The next move belongs to firms and GCs. They have to set clear guardrails, train people to question the machine, and make human oversight a feature of India's legal AI story rather than an afterthought.
August is a legal AI workspace serving law firms and in-house teams in India and globally. It specializes in building personalized Legal AI solutions for teams and also helping to automate repetitive tasks and workflows across organizations.