A powerhouse at the forefront of the legal industry’s AI revolution
Greg Dickason remembers how software used to be built. Not long ago, product teams spent weeks gathering requirements, sketching ideas and debating specifications before writing a single line of code.
Today, thanks to AI, that process has been turned on its head. What once took months can now take hours; a working prototype can be built, shown to a customer and rebuilt before the meeting is over. “It’s no longer simply writing code,” explains Dickason, chief technology officer at LexisNexis Legal & Professional. “It’s understanding what you need to build and getting that in front of customers as fast as possible.”
But in a profession defined by precision and accountability, speed alone proves very little. Being first to market with a product that lacks authoritative content and expertise to validate its output is of little use to a lawyer whose reputation depends on getting the answer right.
That is the distinction LexisNexis, part of RELX, has spent more than a decade building toward. Once best known as a legal research provider, the company has invested more than $1.2bn in its technology stack and built a global organization of more than 4,000 technologists and legal experts. The result is not a single product but an integrated system in which faster development is only valuable because of what sits behind it: a repository of authoritative legal content, a flexible, multi-model platform that draws on the best available AI without being tied to one provider, a development model that brings customers into the process, and a structured human evaluation function that holds every output to the standard that legal work demands. No single component is enough on its own. The advantage comes from connecting them in a way that mirrors how legal professionals actually work, an ecosystem LexisNexis describes as its Legal Intelligence Engine.
Beyond the model: engineering the agent harness
Most conversations about legal AI still center on the model — which one is fastest, which one scores highest on a given benchmark. Inside LexisNexis, that's only half the story. A large language model, however capable, still has to be told what to do, given the right tools at the right moment, and checked before its output reaches a lawyer — and that surrounding structure, not the model alone, is what determines whether an AI agent can be trusted with real legal work. In agentic AI, that surrounding orchestration layer is commonly known as a "harness."
Lawyers don’t experience their work as a stack of AI technologies. They experience a matter – a transaction, a negotiation, a client problem to solve. The harness allows them to state in natural language what they need done, then enables the AI agent to select the right skill, the right model and the right sequence of steps to complete the work. Rather than requiring attorneys to run a series of narrow, task-specific prompts, Protégé is designed to understand the end goal and carry the work forward across a project.
"The effectiveness of an agent comes from the combination," Min Chen, chief AI officer at LexisNexis, explains, "not the model by itself, and not the harness by itself — the model's intelligence and the engineering around it, working together."
That gives LexisNexis two levers for making Protégé smarter: improving the underlying models and improving the harness that orchestrates them. Rather than waiting on the next model release, her team treats the harness itself as the place where quality and reliability actually get won: deciding, in real time, which specialized sub-agent to call for a given task, when to seek human input, and when the model can be trusted to carry a step on its own. As leading models begin to converge in raw capability, LexisNexis sees harness engineering as an increasingly important differentiator — the discipline that determines whether an AI agent can reliably decide which sub-agent to call, when to ask for human input and when it can move a legal workflow forward on its own.
“We were at the forefront of this development because we recognized that customer needs were more complex than prompting could handle,” Chen, who oversees roughly 1,000 data scientists and engineers, says. “Law firms want AI to handle bigger, longer-term projects and we’re working to give them that.”
LexisNexis' Raleigh Tech Center is home to 700 engineers and data scientists
LexisNexis' Raleigh Tech Center is home to 700 engineers and data scientists
For LexisNexis, the reliability requirement is non-negotiable. The harness is built around the same trust boundary that runs through the rest of its AI strategy: authoritative grounding, permissions, auditability and human review stay fixed even as the automated parts of the system get faster and more capable. It's why, Min says, harness engineering isn't a side project sitting next to the model — it's the layer that decides whether everything else LexisNexis has built, from its content to its human experts, actually reaches the lawyer in a form they can rely on.
The harness also gives LexisNexis a way to remain tech-agnostic and integrate the strongest available models rather than betting on any single one. Lexis+ with Protégé draws on both purpose-built legal AI models and leading general AI models from OpenAI, Anthropic and Google, with which LexisNexis has long-standing strategic partnerships.
“We will use the best model for each unique use case to provide the best result,” says Jeff Reihl, LexisNexis’ technology chairman. “Our customers know they get the fastest speed, at the lowest cost, and the highest quality because we’re not forcing a single model to do everything.”
Building with the customer
If harness engineering defines how LexisNexis turns models into trusted legal AI agents, its development model determines how quickly those agents improve. Delivering those capabilities requires continuous experimentation and getting clients to participate, placing them side-by-side with developers so their input is embedded from the outset.
This mindset has also allowed LexisNexis to offer solutions to the common pain points lawyers had as they began taking up AI-assisted legal tools, explains Dickason. For example, some law firms found themselves using a different product for each of the many tasks they undertake; others reported issues with cumbersome prompting.
In addition to this closer cooperation, much faster coding with the help of AI (known as vibe-coding) has further reduced the time for prototyping, often from weeks to days or even hours. “The paradigm has shifted how we build products and how we interact with customers – we can take an idea, build a model, show it to customers and iterate on an actual working prototype,” Dickason says.
That work is increasingly happening face-to-face inside LexisNexis' new Customer Innovation Lab in New York City. There, customers, engineers and product teams work side-by-side to test ideas and refine working prototypes in real time, often within the same session. The goal is to compress the distance between customer feedback and product development, replacing long development cycles with continuous iteration.
"The Customer Innovation Lab changes who's in the room when legal AI gets built," says Dickason. "Customers help originate and shape what we build from day one. Our job is to validate it, ground it in citable legal authority, and decide what's ready to scale into Lexis+ with Protégé, the platform our customers rely on every day.”
By turning customer needs into working solutions quickly enough, the product development becomes a continuous collaboration rather than periodic consultation.
LexisNexis' foundation of trust
is built on
>200bn
legal and news documents and public records
4m
new records
added daily
50k+
sources
3k+
courts provide near real-time updates
The human in the loop
Amid the cutting-edge technology and AI, it is humans who are front and center in LexisNexis’ innovation strategy. Across the firm’s global network, legal experts and subject matter specialists work alongside technologists to help ensure AI outputs meet the standards required for legal work. Whereas large language models, like those used in popular chatbots and search engines, don’t have the level of nuanced legal judgment that trained human experts do, LexisNexis embeds the expert knowledge of attorneys into products through a continuous feedback loop linking design, testing and engineering.
“We have millions of hours of accumulated experience and local legal expertise from thousands of lawyers around the globe – that’s not easily matched,” says Serena Wellen, vice president of product management at LexisNexis. “We couldn’t do without that team because you can’t develop models without knowing if its output is useful to customers.”
Officially part of the Data Discovery and Enrichment (DDE) team but often referred to as the ‘humans in the loop,’ the 250 professionals under Wellen’s leadership ensure that lawyers don’t have to rewrite a deposition or go elsewhere to fact-check a citation. And their task is not a simple one, like choosing between a good answer or a bad one. Instead, it involves applying a spectrum of metrics that reflect the subtleties of courtroom requirements which attorneys must fulfill, including fluency, accuracy and authoritativeness.
“This is really a specialized art, because you need knowledge of the customer, the law, data analysis and a basic understanding of how the overall technology works,” says Wellen.
The team of human experts is divided into broad categories such as workflow, drafting and research, and then again into subcategories such as litigation and transactional drafting, for example. Those units are replicated for major jurisdictions and languages. So, you might have a group of specialists in Australian litigation drafting that evaluates output and provides feedback to data scientists.
That expertise, insight and efficiency allow for faster iteration of data experiments, meaning speedier improvements of AI models. As a result, the length of testing cycles has been reduced from days to hours, allowing for an ongoing improvement loop. Instead of launching new models periodically, LexisNexis constantly updates and enhances its products, giving customers the reassurance of always having the latest and the best.
Authoritative content: a foundational advantage
For all the speed of development and rigor of validation, none of it works without the content beneath it. That content forms the foundation of the broader system LexisNexis has built, connecting authoritative information, technology and human expertise around legal work. The Legal Intelligence Engine is the ecosystem that powers the work.
For lawyers needing to meet courtroom standards, inaccurate content is a huge reputational risk. According to a database compiled by lawyer and data scientist Damien Charlotin, there have now been more than 2,000 cases globally where a court has commented on AI-generated hallucinations in legal filings — and the penalties are growing larger.
LexisNexis' advantage is its own, comprehensive content of more than 200bn legal and news documents and public records from 50,000 sources, with near real-time updates from more than 3,000 courts. This repository is continuously updated with around 4m verified additions every day, with no knowledge cut-off date.
The database includes content that would be impossible to replicate or to find, some of which is no longer available, explains Dickason. So, not only do lawyers avoid erroneous information, but they obtain information they can’t get anywhere else. Content, however, is not the only piece.
“We’ve amassed not only content but technology capabilities that we can now leverage in addition to the newest AI capabilities coming to the table,” Reihl says. The advantage comes from how those pieces work together. The authoritative content, capabilities and customer data are combined in one intuitive experience.
One of those capabilities is retrieval-augmented generation, or RAG, which enables AI models to retrieve and use relevant information from authoritative sources when generating responses. The legal knowledge graph combined with Shepard's Citations provide additional layers of validation and context.
To ensure confidentiality, queries, chat history and uploaded files are never used to train public AI models, and users have a range of options around whether and when to save such data. LexisNexis runs its GenAI within its own secure, encrypted environments.
In practice, a legal professional can describe what they want to accomplish, and Lexis+ with Protégé determines which AI capabilities, tools and trusted sources to apply behind the scenes. That includes carrying context between tasks.
Taken together, it is a combination that customers say is difficult to find elsewhere. “The market is flooded with legal AI companies who make many promises,” says John Ursin, managing partner at Schenck Price, a full-service law firm that has been serving the New Jersey and Manhattan areas for 114 years. “We needed to partner with LexisNexis due to its unmatched relationship with the legal sector and its long track record of providing the very best in legal tools.”
In an area moving as quickly as AI-driven legal tools, that staying power was decisive. “We could not risk partnering with a provider that cannot keep up and whose products might be obsolete in six months,” Ursin says. “I really think that LexisNexis has a multiple-decade head start over other providers.”