Most contracts are written for the parties. The good ones are written for the court.

This is a distinction the Indian legal tradition has always understood, though it is rarely stated aloud. A well-drafted Indian contract carries two layers of communication running in parallel. The first is what the parties have agreed. The second is a set of instructions to any court that might one day read the document. The instructions tell the court how to construe the clauses, which provisions take priority over which, what should be treated as substantive and what as clarification, and where the drafter wants the court to defer and where the drafter wants the court to override.

The instructions are not in footnotes. They sit inside the prose, encoded in a small set of conventions every Indian commercial lawyer has been trained to recognise.

Notwithstanding anything contained in is an instruction. It tells the court to read the clause that follows as overriding any other provision in conflict with it. Subject to the provisions of is an instruction in the opposite direction. It tells the court that the clause that follows yields to the provision named. For the avoidance of doubt is an instruction marking what comes next as clarification, not as a fresh obligation. Without prejudice is an instruction that the communication or admission that follows cannot be used to defeat a right the party has not yet exercised.

These phrases look like style. They are not. They are how the drafter speaks to a court the drafter will never meet, in language the court has agreed to read in a particular way.

The Supreme Court has accepted these conventions as instructions for over six decades. In Chandavarkar Sita Ratna Rao v. Ashalata S. Guram, the Court held that a non-obstante clause is appended with a view to give the enacting part an overriding effect in case of conflict. In South India Corporation, the Court held that subject to conveys the idea of one provision yielding place to another. Parliament uses the same conventions in the statute book. Explanation 2 to Section 34(2)(b) of the Arbitration and Conciliation Act, 1996 opens with for the avoidance of doubt. Section 23 of the Indian Evidence Act, 1872, retained as Section 21 of the Bharatiya Sakshya Adhiniyam, 2023, codifies the without prejudice rule.

When the drafter uses these phrases, the drafter is sending an instruction the court has agreed to receive.

This is the layer that disappears when general-purpose AI rewrites Indian legal text.

A general-purpose model softens notwithstanding anything contained in to despite the other provisions. It softens subject to the provisions of to in line with. It removes for the avoidance of doubt because the clause that follows reads cleanly without it. The resulting prose is more accessible. The contract reads better. The parties understand it more easily. None of that is wrong on its own terms.

What is missing is the instruction layer. The drafter is no longer speaking to the court. The drafter is only speaking to the parties.

This is the failure mode the profession has not yet named. Hallucination is visible. Fabricated citations get caught. Smoothed conventions do not get caught, because they do not look like errors. They look like better writing. The clause reads more naturally. The intent appears preserved. The contract executes without incident. The cost surfaces only when a court eventually reads the document, finds no instructions, and falls back on default rules of construction, which is the position the drafter took the trouble to draft away from.

The Supreme Court has made this cost concrete in the last eighteen months. In Annaya Kocha Shetty v. Laxmibai Narayan Satose, the Court held that the absence of a possession clause in a document styled as a conducting agreement was dispositive. The clause was not there, so the instruction was not there, so the document could not be read as a lease. In Nagreeka Indcon v. Cargocare Logistics, the Court held that an arbitration clause using the word can indicated only a future possibility, not a binding agreement. A single auxiliary verb decided whether an enforceable arbitration agreement existed. In South Delhi Municipal Corporation v. SMS Ltd., the Court advised the profession to draft arbitration clauses with piercing precision and clarity, and warned that ambiguous clauses risk both invalidation and cost orders.

In each case, the court asked what the document said. The court did not ask what the parties might have meant. The court read the words as instructions and enforced what it found.

The instruction layer is the part of Indian drafting that general-purpose AI cannot preserve, because it does not know the layer exists. The model is trained to produce clear, readable English. It is not trained to produce a document that issues legally weighted instructions to an Indian court. Those two outputs look similar on the page. They are not the same document.

The implication for the profession is sharper than it first appears. The risk is not that AI-drafted contracts will be visibly wrong. The risk is that they will look right and quietly stop instructing courts. The contract reads well. The clauses are present. The senior partner reviewing the draft sees nothing to flag. The exposure surfaces years later, in arbitration or in litigation, when a court reads the document and finds no instruction telling it which clause prevails, or which provision yields, or what the parties intended to clarify and what they intended to create.

The Indian drafting tradition has always treated language as instruction. The tools the profession is now adopting have no such tradition. Closing the gap between the two is not a matter of better prompting. It is a matter of building tools that understand what the drafter is doing.

That is the design choice NYAI was built to honour.