AI is now the cook in the eDiscovery kitchen. The real question is who’s tasting the food.
I just got back from Relativity Fest in Chicago, and one message came through in every keynote, session and hallway conversation: AI is no longer an add-on in eDiscovery. It’s the platform. RelativityOne is now Relativity aiR, and Relativity reported more than 740 million “high-value results” in 2026, up from 180 million the year before.
But here’s the part of that number that caught my attention. Relativity defines a high-value result as AI work confirmed by a qualified human. Even the platform measures AI’s value by the person who validates it. That’s the lens I brought home from Chicago.
For context, I’m a Relativity Certified Sales Professional. I’ve worked on thousands of matters over 18 years in this business. I see AI from a 10,000-foot view, with enough time in the trenches to know where I’d train the AI more than it would train me. Here’s what I took away.
Speed without constraint just goes too fast
Much of RelFest centered on simplicity: prompt validation, assisted workflows, and helping case and project management teams keep the human in the loop instead of forcing them to build the loop. The tools now take on classic discovery bottlenecks, such as large data sets, unstructured data, and the balance between precision and recall. And they’re fast. In September alone, aiR Review usage passed all of 2025.
You can now load your complaint, your ESI protocol, and the steps across the EDRM, from identification and preservation through collection, into the tool and let it create some magic on the front end. That sounds great. But speed without constraint just goes too fast.
Would I take a Ferrari to go five miles? Or would I put it on a track, prove it can go 0 to 60 in 3.6 seconds, and then trust it on the road? One session, How to Choose the Right AI Tool for the Job in Litigation Workflows, made the same point: match the tool to the task, and build in quality control, validation, and human oversight. Sometimes the vehicle just needs to get from point A to point B. Proportionality still matters.
Validation is the new first-pass review
Data now routinely reaches terabytes, so the goal is to cut through the junk and get to the richness of the data. Say AI culls 1 million documents down to 15% that likely matter. That’s still 150,000 documents that someone must validate, review for privilege, redact, log, and produce.
AI can’t account for the lawyer, counsel, and team who then work through that set to find what’s key and responsive to the case.
AI is also only as good as the instructions given. Vague criteria produce vague results, and only a reviewer who knows the case will catch a confidently wrong answer.
The courts are paying attention too. This year’s Judicial Panel theme was Preservation, Proportionality, Privilege, and Prompts. A session on rethinking ESI protocols covered how parties should negotiate AI use and validation up front, and another asked outright: Privilege on the Line: Does Using AI Waive It?
The takeaway: validation isn’t an afterthought. It’s a staffed, documented workflow, and it’s where defensibility lives.
The skills gap nobody’s billing for
Junior associates used to get their hands dirty on first- and second-pass review, and service providers handled the mechanics: coding, tagging, redacting, and building privilege logs. Today, AI is the cook in the kitchen.
“ AI is now the cook in the kitchen. But the cook’s work is only as good as the person who knows how to taste it. ”
That’s where expertise in the system is key, and I don’t think skill levels got enough attention. Lawyers practice law. Some of the great ones practice tech. And some of the greatest know law, tech, and AI well enough to outthink the tools themselves.
The Shaping the Next Generation of Legal Professionals session looked at the skills future lawyers will need. But the industry has a nearer-term question: if AI takes over the first pass, where do new reviewers learn the judgment that validation requires?
What makes any system great is that you learn it, then build better workflows inside it based on your own expertise. A session called A Playbook for Building Effective AI Workflows around aiR asked why outcomes vary across matters. My answer: you’re only as great a thinker as the person prompting you to think.
The conversation that wasn’t on the agenda: value vs. the billable hour
I didn’t see a session dedicated to pricing, yet it sits under every AI conversation. What does the end client want? Value, time savings and cost efficiency. What is the lawyer paid by? The billable hour. Every piece of software has a cost, so where does the client see the value?
The honest answer is that it depends on the case. Some matters need the Ferrari. Many just need a reliable car and a skilled driver.
What does this mean for you
- Law firms: aiR’s speed is now table stakes. Your value is the judgment behind it: validation, privilege calls, and a defensible methodology. The challenge is staffing that works without carrying an expensive bench or writing off review hours.
- Corporate legal teams: AI makes the first pass cheaper, but validation, privilege, and QC are often still billed hourly through outside counsel. Ask who validates the AI on your matters, how it’s documented, and what it costs. Predictable pricing and human-verified results should be the standard.
- Review providers and LPOs: Linear review hours are shrinking, but the work is moving, not disappearing. New aiR workspaces this year included internal investigations (20%), FOIA, regulatory requests, data subject access requests, and breach response alongside litigation. The opportunity is in validation, QC, and project management across all of those matter types.
- Teams still on Relativity Server: Relativity confirmed its January 2028 policy for new matters on Server. If you run a large environment, treat 2027 as the year to execute, not the year to start planning.
Where Aeren LPO fits
This is the work we do every day. Aeren LPO’s India-based reviewers and Relativity project managers work inside your Relativity aiR environment to validate AI results, handle privilege and redactions, and QC every AI-flagged set, often overnight, so your team wakes up to finished work. AI does the first pass. We help make it defensible.
Let’s compare notes
Met us at RelFest, or wish you had? I’d love to hear what you took away from Chicago and how your team is handling AI validation.
Sources
- Notable Stats from the RelFest Keynote – eDiscovery Today
- It’s All Here: What’s New in Relativity aiR – Relativity
- RelFest 2026 Tuesday Sessions – eDiscovery Today
- RelFest 2026 Wednesday Sessions – eDiscovery Today
- RelFest 2026 Thursday Sessions – eDiscovery Today
- RelFest 2026 Takeaways: Putting AI to Work and Planning for 2028 – TCDI
