The legal industry is facing a paradox: Never before have the capabilities of artificial intelligence been as impressive as they are today. Within seconds, AI systems draft contracts, analyze case law, summarize thousands of pages, or propose litigation strategies.

Nevertheless, lawyers and corporate legal departments continue to review the documents produced, verify the analyses, and work through the legal reasoning themselves before making decisions. AI is used daily, but only a few are willing to rely on it without reservation.

The decisive question

Why? This question is now more important than which model is the most powerful or has access to the largest data set. If lawyers continue to review every case themselves, it is probably because we have only solved part of the problem.

"In law, a plausible answer is not enough."

In 2017, Ashish Vaswani and his seven co-authors published the scientific paper "Attention Is All You Need." The Transformer architecture introduced in it paved the way for today's large language models and opened up possibilities in language processing and generation that had previously been unimaginable. But this progress does not answer another central question, especially in professions with a high level of decision-making responsibility: How can a machine be enabled to think and argue like an experienced practitioner?

From plausible texts to reliable conclusions

In law, a plausible answer is not enough. An analysis can be linguistically flawless and still contain legally incorrect conclusions. It may rely on facts that are by now outdated, overlook a crucial piece of evidence, or ignore the sequence of events in a case and the consequences of a legal relationship. The challenge, therefore, is to generate legal conclusions in whose reliability professionals can increasingly place their trust.

For decades, lawyers have relied on collections of contracts, pleadings, legal opinions, and reports. Reusing earlier work saved time, reduced risk, and lowered the cost of document production. Today, artificial intelligence can create drafts or summaries within seconds. That raises a new question: Should AI learn what lawyers have done in the past, or how a lawyer should think and argue legally today?

"Human expertise does not consist solely of remembering comparable cases."

These considerations led us to develop the "Zero Legacy" approach. The deliberately provocative term by no means questions the value of earlier documents as evidence, reference material, or records of a law firm's working methods and a client's strategy. But the creation of legal work should in future be guided by methodology: the machine should learn how an experienced lawyer argues, rather than merely reproducing that lawyer's previous work.

Human expertise does not consist solely of remembering comparable cases. With increasing experience, intuition, judgment, and a solid methodology develop. Memory provides context; methodology guides action. So why should artificial intelligence rely exclusively on documented knowledge?

From document knowledge to legal understanding

Today's AI systems are excellent at researching information. However, they still struggle to reconstruct a case in its entirety.

A legal case is a dynamic structure in which facts, parties, contracts, evidence, and legal norms are interconnected and change over time. The lawyer's task is to understand these connections, identify missing information, and assess their legal consequences in order to develop a coherent argument. A trustworthy AI must arrive at a comparable understanding.

We are convinced that the next generation of legal AI will depend less on the sheer performance of the models or the volume of data than on an architecture that can replicate the legal thinking and reasoning of an experienced lawyer.

Our platform is based on four pillars: an explicitly defined legal methodology, a memory that reconstructs the actual relationships within a case, a firm-specific intelligence that remains entirely owned by the respective law firm, is never shared with others, and can be fully exported at any time, and an orchestration layer that connects these components.

Our goal is to capture a legal case as a whole, trace its development, and then apply the appropriate methodology based on that understanding. We also believe that a firm's internal working methods, negotiation positions, preferences in drafting legal texts, and accumulated experience represent a strategic asset. This knowledge must remain confidential and transferable, and belong to those who created it. It must not be used to improve the AI systems of other users of the platform.

"The next revolution will come from systems that professionals can finally trust."

Aviation provides an interesting analogy here. The Wright brothers taught humanity to fly. A few years later, Lawrence Sperry paved the way for the autopilot, which made it possible to delegate certain tasks to the machine. This allowed the pilot to focus more on decisions and handling unforeseen situations.

More than a century later, artificial intelligence is still discussed primarily in terms of "copilots." The challenge now is to develop systems that are reliable enough to gradually take on the role of an autopilot — while responsibility, strategy, and judgment remain with humans.

Large language models have changed the way we work. They have probably solved the first half of the problem. The next revolution will come from systems that professionals can finally trust.

Dominique Lecocq is the founder of Laine Neural Network.

Read all previous texts by our guest authors here.