The new product, Agent Lumi, will be available in a limited test for customers by the end of the year. 

Legal AI is a field that’s crowded with competitors trying to help lawyers reduce the time it takes to read, amend, and negotiate business documents. Harvey, which offers generative AI tools for lawyers and is backed by OpenAI, has generated considerable buzz, for example, while Clio, a tool for summarizing documents and generating bills, closed a $900 million funding round in July.

Luminance was founded in 2015 by Cambridge mathematicians and originally focused on automating the legal review of contracts. Among its founding investors was Mike Lynch, the former Autonomy CEO who was acquitted of fraud charges involving the sale of that company and died earlier this year when a superyacht he was on sunk in a storm off the coast of Sicily.

The company’s earliest customers were law firms, which let the company access some of the legal contracts they had worked to train its AI. Luminance CEO Eleanor Lightbody says that her company’s legal dataset, combined with custom machine learning algorithms, sets the company’s purpose-built AI apart.

Adam Guthrie, a founder and chief technical architect at Luminance, adds that the model’s view into all of a business’s contracts gives it access to “the heart of each company.” 

Other firms have a head start with so-called agents, which are meant to automate decisions and perform tasks without human oversight. Among others, Agent Lumi will go up against AI startup Spellbook, which in August premiered Associate, pitched as “the first fully in-depth agent for legal work.” Like Harvey and many AI tools, Spellbook is built on OpenAI’s foundational model.

Agent Lumi is launching a year and a half after Luminance introduced its chatbot, Ask Lumi. The generative AI model answers business questions and redrafts contract language. 

Earlier this year, Luminance launched an auto markup tool that can review contracts, redraft clauses, and bring outside contracts in line with company standards, but all in isolation. That’s what Lightbody refers to as a “passive” agent. Agent Lumi will be “proactive” and run multiple parallel tasks at a time through a “chain of reasoning,” as Guthrie says.

“All of these tasks—super repetitive, automated tasks—are going to be pieced together to allow us to refocus on the value add or the strategic thinking,” Lightbody says.

In another contrast to more “generalist” models, Luminance stresses that its “legal-grade” AI is trained to prevent hallucinations, or mistakes. Rather than confidently offering incorrect answers like some mainstream models are known to do, Luminance says its model is programmed to defer to human professionals on what it doesn’t know. 

Lightbody says that feature is huge for lawyers, for whom precise and accurate language is essential. 

“They’ve trained our AI into a totally different answer, which is, if you don’t know the answer, tell the human who does know the answer,” Lightbody said. And in an effort to build trust, she says Luminance users will be able to check the agent’s homework. That means letting them audit “every step” that Agent Lumi took, such as where in a contract information was pulled from.

“The more that you do that, the more you can then really hand over the reins in time,” she said.

But while it touts the ability of its new AI assistant, Luminance warns that like any legal tech, it’s just an assistant to human lawyers. If something is incorrect, the human is still liable.

Working with large legal firms early on has been critical for gaining new customers as Luminance expands its business to new markets, Lightbody adds.

“They’re gatekeepers of the data, but also, if a law firm will use our software, then their customers would also use our software,” Lightbody explained.

Today, Luminance is reaching customers beyond the legal profession, increasingly targeting a corporate customer base. The U.K.-based company closed a $40 million Series B funding round in April. The company declined to disclose its valuation in that funding round. Luminance  also moved into the U.S., opening offices in Dallas and ramping up hiring on the West Coast. More than a third of annual revenue is now generated in the U.S., it says. 

Scaling up business operations only improves the quality of the product, Lightbody adds. It’s the variations in the data, not just increasing the size of a static repository of information, that make the technology work.

“The AI knows that a change in control clause might be written 100 different ways, but there might be another way that it’s written,” she explained.