Building AI for the Realities of Legal Work

Authored by Seeta Bodke, Head of Core Product Pacific

Artificial intelligence is rapidly moving from experimentation to everyday legal practice. The question facing legal teams is no longer simply whether lawyers will use AI. Increasingly, it is how AI can be applied in a way that meets the standards legal work demands.

That shift was at the heart of a recent LexisNexis® webinar exploring the evolution of legal AI and the latest developments in Protégé, LexisNexis’ AI-powered legal assistant.

According to the latest LexisNexis APAC AI sentiment survey AI adoption is becoming more embedded in day-today legal work with 73% now using AI tools compared to 60% last year. In Australia, 54% identify time savings as the leading measure of AI value, with faster research, document review and first drafts among the most immediate benefits.

But adoption is only part of the story. As AI becomes embedded in legal work, the focus is shifting from access to trust, context and outcomes.

From answering questions to completing legal work

Much of the first wave of generative AI centred on a relatively simple interaction: enter a prompt and receive an answer.

Agentic AI represents an important evolution. Instead of treating a legal problem as a single question, an agentic system can break the problem into steps, identify the tasks required, retrieve relevant context, undertake drafting or analysis, and iterate as the work progresses.

That distinction matters because meaningful legal work is rarely a one-step activity. Preparing a memo, reviewing a contract, analysing a matter or assessing risk can require multiple stages, sources and decisions.

The opportunity is therefore not about removing lawyers from the process. It is about enabling AI to orchestrate more of the work around them, while lawyers remain responsible for judgment, review and final use.

The legal AI stack: more than a model

This is also why legal AI cannot be defined by the underlying large language model alone.

Protégé is built on a broader legal AI stack designed to bring together the different components required to support legal work.

At its foundation is authoritative legal content. On top of that sits context engineering, including retrieval, knowledge graphs and agentic retrieval-augmented generation (RAG), designed to provide the AI with the appropriate legal context. Agents, workflows and enterprise integrations then help orchestrate tasks using trusted LexisNexis content alongside an organisation's own documents and metadata.

The objective is not simply to generate an answer. It is to support the creation of actual legal work products, from research and contract analysis to memos, briefs, presentations and other outputs within the tools lawyers use.

This stack also reflects three important principles behind Protégé: grounding, traceability and flexibility.

Grounding means legal AI is anchored in verified sources rather than relying primarily on a model's general knowledge. Traceability enables lawyers to inspect the authority behind an answer. And flexibility recognises that legal professionals need different ways to work depending on the task, sometimes asking a question, sometimes running a repeatable skill, and sometimes using a structured workflow to produce a defined work product.

What does that look like in Protégé?

Protégé brings these elements together through several modes of working.

Ad hoc prompting gives lawyers a familiar conversational starting point for research, drafting, summarisation and document analysis. For more complex research, AI-guided research can create a research plan, break a question into subqueries and identify proposed grounding sources including cases, legislation and secondary materials for the lawyer to review before the response is generated.

Workflows provide predefined steps for common types of legal work. In the webinar, a drafting workflow demonstrated how an agent could review uploaded documents, iteratively retrieve and verify information from LexisNexis content, identify gaps and use those sources to produce a more complete draft.

Skills provide another layer for flexible, multi-step work. A skill can dynamically create a plan for completing a task, allow the lawyer to review or modify that plan, and then execute substantial analysis. In one demonstration, a litigation-focused skill analysed court submissions to produce a structured report capturing key facts, dates, submissions, authorities and other details.

Together, these experiences show how the legal AI stack translates into practical legal work: trusted content provides the foundation, agents help plan and execute tasks, and workflows and skills provide structure while keeping the lawyer involved.

Trust remains the foundation

As AI becomes more capable, governance becomes more important, not less.

Legal professionals need to understand the sources an AI system has relied upon, how their information is being handled and where human review is required. The appropriate level of oversight should also reflect the risk of the task: summarising information is different from preparing complex advice or a client-facing document.

That makes human-in-the-loop design, source traceability, security and authoritative legal content fundamental to responsible legal AI. As discussed during the webinar, trusted and up-to-date legal content can help reduce hallucination risk and improve the quality of legal reasoning.

The next phase of legal AI will not be defined simply by which model can generate the most impressive response. It will be defined by how effectively the entire legal AI stack works together, combining models with trusted content, context, agents, skills and workflows to support the realities of legal practice.

That is the direction embodied by Protégé: moving beyond isolated prompts towards AI that can support more complex legal work while giving lawyers the grounding, transparency and