Every year, the Blickstein Group’s Law Department Operations Survey gives legal ops leaders a rare thing: a real benchmark, not just a vendor pitch. This year’s webinar, hosted by Brad Blickstein, explored where legal departments truly stand on AI, compared to where the marketing suggests they should be.
If you run legal operations for a large global organization, here’s what’s worth taking back to your team.
1. "Piloting" is the norm — and that's not a bad thing.
The survey found that almost every legal department has done something with AI, but only a small slice describes themselves as fully operational. Twenty-three percent claim to be “fully operational” but client conversations reveal that true end-to-end integration is rare.
The panel’s read: most organizations are still testing, and that’s appropriate given how fast the tooling is changing. The bigger risk isn’t moving too slowly — it’s equating “we turned a feature on” with “we integrated AI into how the department actually works.” Many tools are being “turned on” but are not woven into holistic processes yet.
Takeaway: Don’t let a vendor’s activation metrics stand in for genuine adoption. Ask whether AI has improved a workflow, not just whether someone clicked “enable.”
2. Usability now beats security as the top concern — and that's a sign of progress, not risk-taking.
For the first time in the survey’s history, usability edged out security as legal ops’ top AI priority. That might sound alarming, but the panel framed it as a natural sequencing issue, not a lapse in judgment.
Usability drives adoption, adoption drives results, results are what we are requiring from AI. If we can’t get users on board and security isn’t going to matter. It only matters if people are actually using the tool.
Hand and hand with security is transparency and reliability. The panel discussed the importance of remembering that real generative based AI still doesn’t give you the same answer the same way every time. That means legal professionals need to know what they are asking for and what types of outputs require critical review. Humans are key to AI security.
Takeaway: If your AI rollout stalls because a tool is technically secure, but nobody wants to use it, you haven’t solved the problem — you’ve just moved it. Build for usability first, then layer in the governance with an eye toward transparency and a good understanding of use cases.
3. Legal is quietly becoming the AI governance hub — whether it planned to or not
One of the more striking data points: roughly a third of legal ops leaders say they now advise on or own AI governance for the entire business, not just the legal department. Historically, legal almost never had that kind of cross-functional reach outside of privacy.
This is a natural connection to data governance — since most existing data governance functions already lived inside legal as a function of privacy or data security. AI oversight naturally landed there, too. Acceptable-use policies gave legal an early, visible seat at the governance table, which became a springboard for broader influence.
Takeaway: If your legal ops function isn’t yet positioned as a cross-functional AI resource, expect that to change. The organizations ahead of the curve are treating legal’s governance experience as a company-wide asset, not a departmental afterthought.
4. "Agentic AI" needs a much sharper definition before you build around it
Nearly every vendor now calls its product “agentic,” which the panel agreed makes survey responses about agentic adoption almost meaningless on their own. A lot of what gets marketed as agentic might just be a robotic workflow and not really a system that makes decisions and pursues a goal.
Discrete task tools, strung-together workflows, and true end-to-end agentic systems are three very different things, and most legal departments are still early in that continuum. Plenty of “agentic” claims are just automation with better branding.
Takeaway: Before evaluating an “agentic” tool, ask what decision it’s making and on what basis. If it’s just executing predefined steps, you may not need (or want) the added complexity and risk of a true agentic system.
5. The real question isn't "can AI do this?" — it's "what are we trying to solve?"
Perhaps the most practical thread running through the whole conversation was a look back to the classic IT conversation — someone asks to use a new tool, and the first question back should always be “what problem are you solving for?” That logic applies just as much to choosing between Claude, ChatGPT, or Copilot as it did to any legacy software decision.
Takeaway: Resist the urge to start with “which AI tool should we buy?” Start with the pain point, then work backward to the right mix of AI, existing software, and plain old process redesign. Not every problem needs an agent — some just need a well-built workflow.
These takeaways are drawn from the panel discussion at the 18th Annual Blickstein Group LDO Survey webinar, featuring Brad Blickstein (Blickstein Group), Diane Homolak (Integreon), Mike Ferrara (FTI Consulting) and Laurie Ehrlich (Icertis).
