Agentic AI performs tasks with limited human interaction through software programs known as agents. By taking action and making decisions, agentic AI differs from generative AI, which focuses more on creating content.
Both generative and agentic AI operate by using large language models, or LLMs, which use machine learning to conduct human-like tasks. Generative AI relies more on humans providing constant input through prompting or prompt engineering.
Law firms have been increasingly adopting generative AI to become more efficient. Now, experts say that agentic AI might be the next innovation to take hold.
"Law firms are clamoring for AI in a way that they've never clamored for any technology," Annie Datesh, the chief innovation officer for Wilson Sonsini Goodrich & Rosati PC, told Law360 Pulse.
In October, Gartner listed agentic AI as its top technology trend of 2025, citing benefits such as a virtual workforce of AI agents to assist, offload and augment the work of humans.
Several legal tech vendors have started offering agentic AI solutions. Genie AI, a legal tech company that offers a drafting tool and recently raised £13.3 million ($16.8 million), has an agentic AI tool that uses the technology to review and edit contracts automatically.
Spellbook, a company known for its artificial intelligence contract drafting and review software, released an AI agent called Spellbook Associate in August and claims it can, for example, prepare full document sets from a term sheet. Luminance, another contract software company, rolled out a tool called Agent Lumi in September to assist with legal work by automatically editing a contract and bringing it in line with company standards.
As agentic AI takes off, some attorneys may wonder how it stands apart from generative AI.
Peter Krakaur, executive director of the consulting firm Agentic Legal Design, told Law360 Pulse that AI agents assume control over processes and make decisions. These AI agents are designed by prompts but are not controlled by humans at every stage.
"There's an art and a science of crafting the prompts to [get] those agents to take actions," Krakaur said in a statement.
Datesh said agentic systems have aspects of generative AI, including some prompt engineering. The difference is that AI agents finish work quicker than an attorney relying solely on prompt engineering.
Generative AI is considered direct-to-model prompting, where it would take a user multiple rounds of prompting to complete a complex task effectively, such as editing a lengthy contract.
"It's manual, it's frustrating," Datesh said. "No lawyer that I know is going to spend the time to do that."
By contrast, AI agents conduct the work faster and can create more accurate outputs, according to Datesh. This is because generative AI relies more on statistical models that could be more prone to errors.
Datesh said there are multiple AI agents in an agentic AI system coordinating with each other. For example, there might be an AI agent examining the input while another AI agent looks at formatting and another one reviews references. Each AI agent could be programmed to do different tasks, including checking the iteration of work done by other AI agents.
Similar to generative AI, agentic AI faces challenges such as untrustworthy outputs and data privacy. There's also an additional risk with agentic AI, as the automation in the backend could be brittle and break down, according to Datesh. This is because agentic systems might be sensitive to tweaks in their training data.
Experts say law firms should hire data scientists and develop strong guardrails to handle potential complications.
What Agentic AI in Law Firms Could Look Like
Agentic AI may not be widely adopted in law firms immediately, but Krakaur notes that BigLaw firms are more likely to implement AI agents due to their resources. That's why Krakaur, a former EY executive and BigLaw knowledge chief, recently launched his consulting practice to help firms develop AI agent strategies.
Some law firms may be tempted to apply agentic AI directly to the practice of law, but Krakaur advises that they begin by focusing on the business of law first. This includes using AI agents for finding clients and responding to requests for proposals.
"That's where there's so much cost involved and huge upsides for law firms," Krakaur said. "That's where I see the true transformation happening."
However, some BigLaw firms are starting to test agentic AI in the practice of law.
In May, Wilson Sonsini introduced an agentic AI commercial contracting offering for cloud services companies in its Neuron platform. The technology powering the offering is from an agentic AI vendor called Dioptra.ai.
The tool, which is customized for Wilson Sonsini, can mark up cloud services agreements and nondisclosure agreements through a proprietary playbook developed by the firm. In its initial testing, the firm found the tool achieved over 90% accuracy.
Wilson Sonsini is also using the tool to train its attorneys, who ultimately review each completed contract after the markup.
Datesh said the firm trained the tool on synthetic data to protect client data. While the firm may in the future use real client data on the already trained tool, Datesh said there are still concerns about potential regulations from the American Bar Association that need to be ironed out first.
Krakaur said that firms focused on the best data practices are ahead of the game and more likely to try agentic AI before others.
He said firms interested in exploring agentic AI should first develop a clear data strategy and train all their lawyers in AI best practices, because even AI agents need some human intervention.
"We're still in the world of augmented intelligence," Krakaur said. "Everybody is looking towards true artificial intelligence. We're still years away despite what others may say."
--Editing by Robert Rudinger.
For a reprint of this article, please contact reprints@law360.com.
