A recent ILTA webinar conducted by Syllo discussed the impacts of agentic AI in litigation and how it differs from other forms of review.
The panel was moderated by Jeffrey Chivers, CEO of Syllo, and included Melissa Dalziel, Counsel at Alston & Bird; Melissa Fu, Associate at Quinn Emanuel Urquhart & Sullivan LLP, and Dan Stromberg, Partner and eDiscovery Counsel at Outten & Golden LLP.
The discussion centered on agentic AI’s transformative impact on litigation, contrasting it with traditional methods like linear review and earlier versions of Technology Assisted Review (TAR). The panelists shared personal experiences with agentic AI in litigation, highlighting how agentic AI significantly enhances efficiency, accuracy, and strategic advantages throughout the litigation lifecycle, from early case assessment to deposition preparation.
Defining Agentic AI and its Core Functionality
Chivers described agentic AI as a recursive loop to achieve an objective. This system is given that objective and then cycles through several steps, as follows:
- Choosing actions: Deciding what steps to take.
- Performing an action: Executing the chosen step.
- Making observations: Gathering information from the action.
- Updating a knowledge base: Incorporating new information.
- Evaluating potential next actions: Deciding the subsequent steps based on observations and updated knowledge.
This graphic from their presentation illustrates that recursive loop.
When specifically applied to document review in large-scale litigation investigation, this loop involves:
- Evaluating.
- Choosing an action.
- Analyzing documents.
- Performing quality control based on the analysis.
- Extracting learnings from documents and synthesizing them.
- Adding synthesized learnings to a knowledge base.
- Evaluating what to do next.
This graphic from their presentation illustrates that document review loop.
Key Differentiators from Other Document Review Technologies
The panel discussed how agentic AI represents a significant evolution from older methods, including linear review, TAR 1.0, and TAR 2.0/CAL (Continuous Active Learning).
- No Training Seed Sets (Human Labeling): Unlike TAR 1.0/2.0 which require lawyers to review and label documents to train the system, agentic AI does not involve this initial training based on a seed set. Dalziel stated, “Agentic review really is different in terms of the front up burden on the attorneys, which gives you a lot of flexibility in terms of how to use it at various stages of the litigation.”
- Explanations vs. Scores: Agentic AI provides explanations for why a document is relevant, rather than just a numerical relevance score. This offers greater transparency and understanding.
- Handling Nuance and Unlimited Issue Coding: Agentic AI can handle more nuanced issues and allows for “an unlimited number of issue coding across the documents in a cost-effective way.”
- Adaptability: Attorneys can “pivot more flexibly” as new issues emerge in a case, a critical advantage in dynamic litigation.
- Predictability and Defensibility: Stroberg contrasted agentic AI with traditional search terms and TAR 2.0. While search terms offer predictability in population size (but often miss relevant documents), and TAR 2.0 offers defensibility (but unpredictability in review volume), agentic AI offers “both. There’s predictability in time. You don’t know how many responsive documents there are, but it doesn’t matter. It’ll still be very quick and it’ll be defensible.”
Transformative Impact Across Litigation Stages
When it comes to agentic AI in litigation, it can significantly impact the entire litigation life cycle, from early case assessment (ECA) to trial preparation.
- Collapsing Stages: The panel claimed that some stages can be “collapsed,” allowing for simultaneous issue coding and responsiveness review, bypassing traditional data set calling and ECA. In fact, Stroberg boldly stated, “I think ECA early case assessment as we know it is dead,” because agentic AI allows for a more comprehensive and actionable assessment of the entire document corpus much earlier, akin to “complete cases assessment.”
- Enhanced Strategic Insight: Dalziel discussed its utility in initial disclosures, interrogatory responses, and particularly deposition and trial preparation. It can “create a computer brain that mimics my human brain and my understanding the case by reading everything I have read in the case.” This ability to “understand the case as I understand the case” allows it to identify key documents for various stages.
- Rapid Fact Extraction and Chronology Building: The AI can quickly identify key witnesses, documents, and build chronologies, freeing up lawyers for strategic work.
- Identifying Production Deficiencies: Fu described using Syllo to “identify deficiencies in the matter and to come up with a motion to compel in the same time,” even within a 24-to-48-hour turnaround.
Real-World Applications and Success Stories
Perhaps the most compelling part of the discussion was where the panelists shared real-world examples of significant advantages they have personally experienced with agentic AI in litigation:
- Dalziel’s Experience: Faced with amended pleadings close to the close of discovery involving 2 million documents in late 2023, her team needed to pivot quickly. By “giving it the entire case knowledge,” Syllo identified 150 “hot documents” that structured the story, supported the client’s view, and revealed missed significant documents. This led to reliance on Syllo for creating “witness kits for deposition” with 150 documents per witness, dramatically reducing prep time and cost.
- Fu’s Experience: She used Syllo early on with a “relatively simple fact pattern with just like two main individuals”, which still involved a “huge massive volume of documents” but with “a very lean team.” It enabled depositions within weeks, distilling millions of documents into 150 binders. In one case (with 6 weeks to trial, 30,000 outgoing documents and 40,000 incoming documents), Syllo facilitated “overnight review,” identified hot documents, and provided rationales, helping the team learn the case. It also flagged missing information and developed “five motions within two week period.” Fu stated: “Agentic AI was teaching us, was being the superpower behind the human review.”
- Stroberg’s Experience: Faced with 100,000 documents, 80+ requests for production, and only two weeks to review for a client with “very low richness” (i.e., with most documents irrelevant), traditional methods were not practical to meet that time frame. Agentic AI provided review results in “basically a day,” which were “far more accurate in terms of recall than any other approach.” It also delivered “actionable issue codes selected with ration, excerpts, etc.,” something traditional TAR wouldn’t provide.
The panel also referenced Syllo’s compelling white paper titled Agentic AI Document Review Is Transformative for Complex Litigation – available here and also in PDF form here – and covered by us here. It includes ten real world examples involving seven different firms, including those from Dalziel, Fu and Stroberg.
Building Trust and Overcoming Adoption Hurdles
While lawyers are traditionally slow to adopt new technologies, the panelists were optimistic about agentic AI’s faster adoption compared to earlier TAR.
- Self-Validating Output: Dalziel stated: “the output is self validating… the output of the AI validates that it knows what it’s doing [and] that it is providing excellent results.” The provision of rationales and summaries helps users “see that it understands the issues of the case and is able to organize that data by issue.”
- Strategic Advantage and Cost Savings: Dalziel also stated that stated that gaining a strategic advantage and winning a case at a lower cost point is “really addictive”.
- Client-Driven Demand: Stroberg suggested that the widespread public exposure to generative AI will drive clients to ask: “Isn’t there a GenAI thing we could do for this?”
- Security and Privacy Concerns: Dalziel discussed the importance of rigorous security assessments and due diligence to assure clients that data remains within a “little universe” and is not used to train public models. The learning process within agentic AI systems builds knowledge without updating neural network weights, addressing a key client concern.
Impact on Legal Professionals and Work Quality
What about job security in the agentic AI age? The panelists largely dismissed fears of job displacement, arguing that agentic AI instead elevates the quality of legal work.
- Smarter Lawyers, Higher-Level Work: Dalziel stated: “it really is just requiring all of the lawyers to operate at a higher level and do more sophisticated work.” Chivers noted that it “forces the case teams to be more sophisticated…and sharp in their thinking with respect to what matters earlier.”
- Focus on Strategy: Fu stated that agentic AI makes lawyers smarter and enables them to “spend more quality time using my billable hour to help the client”, focusing on “higher level strategic work” rather than manual document review.
- Consistent Review and Bird’s Eye View: Fu also noted that agentic AI provides consistent review – a challenge to accomplish with large human teams – and offers a “bird’s eye view of all potential documents,” aiding in litigation strategy reform.
- Learning and Development: Fu, as a mid-level associate, feels it makes her “smarter and a better lawyer,” providing insights into document importance and helping her learn cases quickly.
Impact on Costs
The panelists also discussed agentic AI’s impact on cost savings, which has become a “100% yes” reality for Dalziel. She stated that she has seen savings achieved through:
- Reduction in Reviewed Documents: AI effectively identifies non-relevant documents, reducing the overall volume.
- Improved Reviewer Pace and Quality: Highlighted sections, rationales, and predicted issue codes speed up human analysis and improve consistency.
- Less Quality Control (QC): The panel stated that improved quality and consistency from AI-assisted review reduces the need for extensive human QC.
Conclusion
In summary, the panel discussed how agentic AI in litigation is disruptive and provides several benefits. Its ability to learn and adapt, provide contextual explanations, and operate with speed and accuracy can fundamentally change the approach to document review, offering significant strategic advantages and demonstrable cost savings while elevating the role of legal professionals.
So, what do you think? What do you think the role of agentic AI in litigation is today? Please share any comments you might have or if you’d like to know more about a particular topic.
Disclosure: Syllo is an Educational Partner and sponsor of eDiscovery Today
Disclaimer: The views represented herein are exclusively the views of the author, and do not necessarily represent the views held by my employer, my partners or my clients. eDiscovery Today is made available solely for educational purposes to provide general information about general eDiscovery principles and not to provide specific legal advice applicable to any particular circumstance. eDiscovery Today should not be used as a substitute for competent legal advice from a lawyer you have retained and who has agreed to represent you.