What if AI were first and foremost a way to read better

There is a lot of talk about generative AI as a drafting assistant. Less as an outstanding reader. It is not (only) a matter of marginal productivity, of a few minutes saved here and there. It is a deeper shift: the ability to digest, cross-reference and prioritise masses of information that no human could absorb alone. Entire functions of analysis, monitoring and preparation are being redefined — not automated as they were, but transformed by tools that bear no resemblance to their predecessors.

In a law firm, this shift is not immediately visible. You do not replace a lawyer with a prompt. But something else is starting to show: legal professionals who read faster, who are better at extracting the useful signals, who cross-reference heterogeneous sources to form a sharper opinion. Where others still open their five tabs of legal commentary, a model has already read, compared and synthesised them. It is another way of working, more fluid, more precise and faster. And it is because the tool does not “do everything on its own” that it becomes useful: it prepares the ground, it sharpens the analysis, it allows the legal professional to bring into play what gives them their value.

Let us take a very concrete example. You receive a 150-page report to analyse urgently — a due diligence, an expert report, an opinion from a public authority. Traditionally, you would have to spend hours on it, annotating, summarising, structuring. Today, some tools make it possible to extract the main points, identify the points requiring attention, and even cross-reference this content with other sources. In a few minutes, you have a structured overview you can query. It is impressive. But it is also the moment to stay alert: have we properly understood what the tool did not read? What did it rely on to reach its conclusion? What biases or errors crept into this quick synthesis? AI gives the impression of having read everything. In reality, it has selected, filtered, interpreted — often without explanation. This is where the legal professional’s role comes back into its own: checking, nuancing, cross-checking, rephrasing.

The point is not to be distrustful on principle, but to work differently. To limit the risks, you need to learn to “steer” AI methodically. First, by structuring your requests: do not ask directly for an answer, but for a process. For example: “summarise the parties’ arguments for me”, then “what are they based on?”, or “can you quote the passages of the report that ground this conclusion?” Next, by testing any blind spots: “which points are missing or under-treated in this text?”, “what alternative interpretations could be considered?” Finally, by cross-checking: never a single answer, always verification by other means. It is not a complex art, it is a (new) discipline. And like any discipline, it is cultivated.

For example, what makes a conversational agent strong is not the first answer. It is the conversation. It is in the back-and-forth, in the rephrasing, in the objection that the tool becomes truly useful. Asking it a question is good. Pressing it again, pushing it to justify, to refine, even to doubt, that is where the real value emerges. This dynamic resembles strategic dialogue more than automatic delegation. You do not use AI like a search engine; you use it as a partial interlocutor, sometimes vague, but incredibly fast and available. In that sense, conversational AI does not distance the lawyer from their reasoning. On the contrary, it can help sharpen it — provided one keeps an active, critical and somewhat playful relationship with it.

To go further, some choose to create assistants dedicated to these complex tasks. A preconfigured conversational agent, trained to analyse a report, answer precise questions, cite its sources, rephrase or flag the limits of a line of reasoning. It is a bit like having a specialised associate, always available, always structured. By feeding it your methodological instructions — set out the context first, isolate the arguments, ask for justifications, flag the grey areas — you turn a generic tool into an operational ally. It does not replace your judgement, but it frames the reading, saves you time and, above all, pushes you to think more clearly. A well-designed “analysis assistant” can be called upon at any moment, in a familiar interface, and carry out its task without drifting. The gain is real. The rigour, intact.

AI is not going to replace legal professionals. It is going to redefine the way they read, reason and decide. To those who see it as a drafting gadget, one can reply that it is far more formidable — and more useful — when it helps us understand better. It is not simply a question of efficiency. It is a question of stance: choosing to remain in control of one’s analysis, to steer the tools demandingly, to cultivate exchange rather than dependence. Reading differently does not mean reading less. It means reading with more discernment, more method, and perhaps — in time — with a little more freedom.