Introducing Agentic AI in Law

You receive a 150-page agreement with a request from the client: “Can you review this and tell me what we should push back on before tomorrow’s call?”

The work that follows is rarely one task. You check the schedules, compare clauses against your playbook, identify deviations, research an unfamiliar legal issue, assess the risks, suggest revisions and prepare your negotiation notes.

AI can already help with each of these tasks. But what if, instead of prompting AI at every step, you could give it the objective and have it coordinate the work from there?

That is the shift from AI assistants to Agentic AI: from helping with individual tasks to coordinating the work needed to achieve an outcome.

The distinction: An AI assistant performs the task a lawyer asks it to do, one prompt at a time. The lawyer decides what comes next and carries the context forward. An AI agent can take a lawyer-defined objective, coordinate the steps needed to achieve it, adapt as it finds new information, and bring the lawyer in at defined review points.

The legal profession has adopted artificial intelligence faster than many expected. Lawyers now use AI assistants to summarise judgments, draft clauses, explain legal principles, translate documents and prepare correspondence. These tools have already made individual tasks faster.

But most legal matters are not a series of isolated tasks. They are connected processes, where the outcome of one step often determines what happens next.

A contract review, for example, may involve checking whether the agreement is complete, comparing it against an approved playbook, identifying deviations, assessing legal and commercial risks, researching relevant law and proposing revised language.

Litigation follows a similar pattern; reviewing pleadings and annexures, identifying evidence gaps, building a chronology, researching authorities and preparing submissions. Compliance work may involve identifying applicable rules, mapping obligations, reviewing evidence, recording gaps and recommending corrective action.

In each case, the work follows a familiar structure, but what happens next depends on what is found along the way. AI assistants can help with individual tasks, but the lawyer still has to connect the dots, carry context from one task to the next, decide what comes next and ensure nothing is missed.

This is where Agentic AI changes the way AI can support legal work. Instead of waiting for the lawyer to prompt it at every step, it can take an objective, plan and coordinate the authorised tasks required to achieve it, preserve context along the way, and bring the lawyer in when professional judgment is required.

Agentic AI starts with an objective rather than a series of step-by-step instructions. A lawyer might ask it to prepare a contract for negotiation, identify gaps in a litigation matter or assess compliance with a new regulation. The system then works out and coordinates the steps needed to achieve that objective, within the boundaries set by the lawyer.

Plans the work. It breaks the objective into individual tasks, works out what needs to happen first and adjusts the plan as new information emerges.

Uses authorised tools and sources. It can work across permitted resources, such as a document repository, clause library, legal research platform or matter-management system.

Carries context forward. It remembers relevant facts, findings, decisions and open questions as it moves from one task to the next, rather than starting afresh with every prompt.

Adapts to what it finds. If information is missing, inconsistent or raises a new issue, it can adjust the next step instead of simply following a fixed sequence.

Works within defined boundaries. It acts only within the permissions it has been given, pauses when approval is required and brings the lawyer in where professional judgment is needed.

This is what distinguishes an AI agent from both a chatbot and traditional automation. A chatbot waits for the next prompt. Traditional automation follows a predefined sequence. An AI agent can determine what authorised step should come next and adapt as the work progresses, while the lawyer remains in control.

Legal work follows processes, but it is rarely mechanical. There may be familiar steps, but what happens next often depends on the facts, the level of risk and the lawyer’s judgment. This combination of structure and judgment makes legal work well suited to Agentic AI.

The key is controlled autonomy. An AI agent can handle lower-risk tasks such as organising documents, extracting dates, comparing clauses or building a chronology. But when a decision involves legal advice, external communication, settlement positions or significant risk, the lawyer steps in.

This allows AI to handle more of the coordination around legal work, moving information between tasks, keeping track of what has been completed, identifying missing information and bringing the work together for review. The AI manages more of the process; the lawyer remains responsible for the judgment.

Lawyers are trained to solve legal problems. Yet a significant part of their time is spent reviewing documents, building timelines, comparing versions, finding authorities, preparing internal notes and assembling final deliverables.

Agentic AI can take on more of this structured, repetitive work and coordinate the steps between tasks. This gives lawyers more time for the work where their expertise matters most; i.e. strategy, client advice, negotiation, advocacy and decision-making.

Legal technology has always changed how lawyers work without removing the need for legal expertise. Online research made authorities easier to find. Document-management systems made information easier to organise and retrieve. Practice-management software made matters easier to coordinate. Agentic AI is another step in that evolution. It can execute and coordinate more of the process, but professional responsibility remains with the lawyer.

For legal professionals, this creates an opportunity to rethink how work moves from instruction to outcome: what can AI handle within defined rules, where must the lawyer approve the next step, and what should remain entirely with the lawyer.

Agentic AI does not mean unsupervised AI. It means giving AI greater responsibility for the process, while keeping legal judgment and accountability with the lawyer.

Deepak Kapoor is the CEO & Founder at Manupatra.