Just a few years after the destructive potential of generative artificial intelligence to hit legal technology, a new subset is emerging: AI agents. This software can perform tasks, make decisions, solve problems, and solve problems on its own with less human guidance than the generator AI. It also enables further cost-saving and speed-up processes.
At least that’s the promise.
The company’s internal legal department was an early adopter of Generator AI. They investigated opportunities for further automation of standard legal tasks, from contract reviews to regulatory-compliant policy checks.
Currently, Agent AI is touted as the next big step to stealing humans from legal tasks thanks to its ability to implement multi-step processes.
Ryan O’Leary, a legal technical expert at research firm IDC, said the company’s legal team is overwhelmed by the tasks they have to undertake. “If we can use (agent AI) to take away the drooping fruit and administrative things, that would certainly be a big win for the organization.” But achieving more and making it faster depends to some degree on the legal department. “Leave AI to your own devices.” O’Leary probably has less human surveillance than generative AI.
So far, AI agents have been very limited, with fewer than 1% of the world’s largest 5,000 companies using it for legal work, estimated Weston Wicks, a legal and compliance technology expert at Research Company Gartner.
So far, legal agent AI software suppliers include US-based legal AI business Harvey and Start-Up Eudia and Sweden’s Legora.
Salesforce, an international business software company that employs around 500 people on its global legal team, handles many legal tasks using its proprietary Agent AI Software, Agent Force, and agent tools from other suppliers. These include answering sales staff questions about customer contracts, negotiating contracts, and developing inquiries to prioritize them before submitting them to the legal team.
The company says its lawyers always check the work of Agent AI.
Nevertheless, Salesforce adds that it can use agent AI software to save around 9,500 hours a year in staff time for compliance and risk tasks. In a pilot project, it turns out that using it to negotiate a non-disclosure agreement is, for example, 25% faster.
If the AI ​​agent is unable to answer legal queries immediately, route them to the appropriate Salesforce lawyer depending on your priority.
“This is to this manager who is assigning work to the right team (takes this),” says Sabastian Niles, Chief Legal Officer at Salesforce.
Ultimately, he speculates that Salesforce AI agents may be able to negotiate part of their contract with the customer or supplier AI agents.
How accurate is the output of Agent AI when used by an in-house lawyer? Niles has no numbers and the company is “very focused” on AgentForce’s accuracy, adding that it is based on a reliable source of data.
AgentForce is used by thousands of companies, including Jobs Marketplace, Tyre Company Goodyear, and Drug Company Pfizer.
Generally, who is responsible for any mistakes that occur?
The answer may be “customers.” Harvey said: “We advise every prospect and client at Harvey that our approach is tech-ready but human-verified, so the final output produced needs to be reviewed thoughtfully by a lawyer.”
The legal department is fully aware of the need for human surveillance by generating the risk of AI hallucinations for agents, sometimes inaccurate information.
Certainly, they may be reluctant to grant AI agents full autonomy, whether they are purchased or built in-house, even relatively basic tasks.
In July, early stage startup Eudia recruited a large team of human legal experts to act as quality control for Agent AI Software when it acquired Johnson Hana, which employs more than 300 people.
“What we discovered is that professionals with ()Agent AI solutions are always better than experts without that solution, or better than that solution without that solution,” says co-founder and CEO Omar Haroun. Eudia’s clients include Battery Business Duracell and Agriculture Group Cargill.
Other suppliers have given examples of how agent AI technology has become more accurate when working without human intervention.
Max Junestrand, co-founder and CEO of Legora, recalls how one client (“market-leading, global financial services company”) caused inconsistencies in the results when their own lawyers and Legora AI agents analyzed old mergers and acquisition transactions. “(It) was really worried that the outcomes would be different to what humans did at Legora,” says Jeunestland. “I realized that humans were wrong.”
The international division of Japanese advertising group Dentsu uses agent AI provided by Harvey.
Robert Clark, general counsel at Dentsu International, oversees a legal team of approximately 260 people except in Japan. This year, they used agent software to ensure that legal policies are compliant with laws and regulations.
The team has not yet calculated the return on investment, but Clark is cautiously optimistic.
One future use of Agent AI is to analyze spending to external legal counsel, for example.
He adds: “(Agent AI) can be transformative, but I don’t think it will be transformative overnight.
