On September 3, 2026, LawNext reported that ILTACON’s busiest news cycle in years centered on two ideas: agentic AI and the Model Context Protocol. Inside that torrent, one update stood out for e-discovery teams: DISCO’s push to remove prompt engineering from AI-assisted review (LawNext).
What changed at ILTACON e-discovery
According to LawNext, DISCO enhanced its AI-powered Auto Review so lawyers can refine instructions without prompt-engineering skills. The system improves tag descriptions from plain-language feedback, runs tests on sample documents, and generates fresh prompts with performance metrics. In short, it turns iteration into a guided loop for non-technical reviewers.
The same LawNext roundup says ILTACON news spanned major e-discovery vendors, including Everlaw, Nuix, Relativity, and Reveal, yet the thread tying them together was clear: agentic AI did the heavy lifting, and standards like the Model Context Protocol (MCP) showed up as plumbing for tool use. That pattern tracks with ILTACON’s broader agenda as the International Legal Technology Association’s annual conference (ILTACON).
Why “promptless” AI review changes budgets and roles
The immediate takeaway isn’t the feature list. It’s the shift in who controls quality and cost. Historically, document review has been the largest slice of discovery spend; prior studies found review dominated total costs in many matters (RAND). When an AI tool lets case teams iterate with natural language, less time gets burned in translation—between litigators, project managers, and external service providers—just to tune prompts.
That opens two changes. First, in-house litigation teams can keep iteration cycles closer to the merits, because the people who know the issues best can now steer the model directly. Second, review vendors will have to justify hours on strategy, sampling, and validation, instead of babysitting syntax or inventing bespoke prompt recipes. The spend moves from technical wrangling to measurable outcomes.
DISCO’s approach—plain-language feedback plus built-in tests—also strengthens the audit trail. If the software tracks how tag descriptions evolved and which sample sets drove metric gains, counsel can explain the process in meet-and-confers and, if needed, in court. That supports proportionality arguments and reduces the risk that the AI looks like a black box. The Sedona Conference’s guidance on technology-assisted review has long stressed transparency around process and validation; integrated testing makes that easier to show without a special report (Sedona).
For law firms, the staffing picture adjusts too. Review managers will spend more time designing defensible sampling protocols, setting acceptance thresholds, and monitoring error drift. Fewer hours should go to wrestling with prompt phrasing or hand-tuning tags. Corporate clients, meanwhile, can push for pricing that links fees to validated outcomes—stability in precision/recall over time, not the number of prompt “experiments.”
Agentic AI and the Model Context Protocol become workflow, not hype
LawNext said DISCO also launched Advanced Research, an agentic AI app for fact investigation that uses multi-step reasoning. That’s the same pattern surfacing across ILTACON e-discovery news: smaller, goal-seeking agents that plan tasks, call tools, and report back with evidence. The move from single-shot prompts to iterative agents matters because discovery tasks often depend on chained steps—identify issues, test a hypothesis, sample targeted documents, then refine.
Standards like the Model Context Protocol fit that arc. MCP defines a simple way for AI systems to request tools, share context, and log interactions. In legal settings, that consistency can reduce integration work and improve defensibility, since tool calls—searches, exports, privilege checks—can be recorded with the same structure across matters. ILTACON e-discovery vendors adopting MCP are signaling they want agents to interoperate and their audit trails to travel with the work, not stay locked in a single interface.
The market read: agentic features will spread first where narrow goals meet repeatable tool calls—email clusters, privilege sweeps, custodian-specific issue tags—before expanding to more subjective calls. Buyers should expect faster iteration on these “thin-slice” use cases, while policies catch up for higher-risk steps.
How legal teams can prepare for AI document review now
Teams don’t need to wait for a full platform shift to capture value. A few moves pay off across tools:
- Define target metrics before you start. Pick precision/recall goals and a sampling cadence you can defend in writing, then hold vendors to them.
- Use seeded validation sets. Keep a stable test pool to quantify improvement when changing tag descriptions or feedback strategies.
- Log every iteration. If the platform doesn’t do it automatically, document feedback changes, sample sizes, and outcomes. You’ll need this for meet-and-confers.
- Price to outcomes, not prompts. Negotiate fees tied to validated performance over time, with incentives for stability across custodians and issue codes.
- Set guardrails for privilege and PII. Even with strong automation, require human-in-the-loop checks for high-risk tags and any export outside the review tool.
This is also the moment to revisit TAR playbooks. Many policies were written for first-generation workflows. Update them to reflect agentic loops, natural-language feedback, and standard tool-call logging. That alignment reduces friction when a court asks how your process maps to established practice—and helps your team compare offerings in an apples-to-apples way.
The upshot for ILTACON e-discovery buyers
The LawNext roundup underscores a shift: vendors are racing to make AI review feel like giving instructions to a colleague, then proving it worked with built-in tests. In practice, that moves spend toward strategy and validation, trims translation costs, and strengthens the record you bring to court. Expect selection criteria to change with it. Ask less about prompt tricks, more about audit trails, metric stability, and MCP support across your stack.
ILTACON e-discovery has often previewed where litigation tech is heading. This year’s push toward promptless control and agentic workflows says the quiet part out loud: the best AI is the kind you can direct in plain English, explain later with numbers, and swap into your process without breaking your audit trail. That’s what buyers should test for next. For more on this, see bloomberg.com and nytimes.com.