For a while, the question was all about what AI could do. Firms ran the demonstrations, tested the tools, and watched the capabilities improve month over month. That question has largely been answered. The one that matters now is harder:
How do you make AI useful every day?
Most firms have moved past the “should we” stage. They have licenses, pilots, and in many cases more than one tool live at once. But access isn't adoption. The real work, the work that determines whether any of this pays off – is figuring out where AI fits into the work lawyers already do, and how it becomes something people reach for, rather than another icon on the desktop that gets opened once and forgotten.
That gap between having AI and using AI is where most firms are stuck right now. And it's rarely a model problem.
The bottleneck isn't the model. It's the knowledge underneath it
AI can only work with the information available to it. If valuable knowledge is hard to find, out of date, or scattered across practice groups, matter management systems, shared drives, and someone's inbox, AI has less to work with, and a more powerful model doesn't fix that. It just makes confident use of incomplete information, which is arguably worse than no information at all.
This is the uncomfortable truth firms are running into. The same barriers that make it difficult for a junior associate to find the right precedent, the right past-matter answer, or the right subject matter expert are the exact barriers that make it difficult for AI to do the same thing. AI doesn't route around bad knowledge management. It inherits it.
So the firms seeing real traction with AI right now are not necessarily the ones with the newest platform. They're the ones that have done the less glamorous work of understanding what knowledge they have, where it lives, how current it is, and who's allowed to use it. That's KM work. It always has been. AI has just raised the stakes and visibility. It's the same structural transformation debt we've seen before, just showing up in the knowledge layer instead of the data layer.
Access isn't the same as openness
Here's where it gets more complicated, and where a lot of well-intentioned AI rollouts stall. Easier access to knowledge must come with the same governance firms already apply to who sees it and when.
Law firms operate under constraints that most industries don't: client confidentiality, ethical walls, outside counsel guidelines, jurisdictional variation, and internal governance built up over decades for good reason. AI doesn't get a pass on any of that. If anything, it sharpens every one of these questions:
- Who should be able to see a given piece of work product?
- When should they be able to see it – during a live matter, after it closes, only with permission?
- How does a firm surface institutional knowledge broadly enough to be useful, without surfacing it so broadly that it creates risk?
- How does governance stay intact when the thing doing the searching is a system, not a person following a policy they were trained on?
These aren't hypothetical. They're the questions GCs, CLOs, risk committees, and CIOs are already asking before they'll sign off on wider rollout. A firm that can't answer these questions confidently isn't ready to scale AI past a pilot, no matter how good the demo looked.
Trust is the actual adoption metric
It's tempting to measure AI adoption by license counts or query volume. The metric that matters is trust. Do lawyers believe the answer AI gives them is accurate, current, and appropriate for them to see?
If the answer is no, adoption stalls. Lawyers are trained skeptics by profession. A single bad or stale answer from an AI tool doesn’t get chalked up to “it's still learning.” It gets chalked up to “this doesn't work,” and the tool quietly stops getting used, even if the firm's usage dashboard says otherwise.
That's why the foundation matters more than the feature set. Confidence in an AI tool is built the same way confidence in a junior colleague is built, through consistent, reliable, well-sourced answers over time. Firms that treat this as a change management and knowledge quality problem, not just a procurement decision, are the ones building that confidence.
Choosing a platform is the beginning, not the milestone
Selecting an AI platform is the first of many decisions, and often the easiest one.
The harder decisions come next. Where does AI sit in a lawyer's workflow: inside the document management system, inside email, as a standalone chat interface, embedded in matter intake? Does it support the way lawyers already work, or does it ask them to change their behavior to accommodate the tool? What happens when the AI-generated answer conflicts with what a partner remembers from a matter three years ago?
The strongest implementations we’ve seen share one trait: they remove friction without introducing new friction in its place. That sounds obvious, but it’s rare. Plenty of AI rollouts solve one problem (finding a document) while creating a new one (verifying whether the document is even current, or whether the lawyer had permission to see it).
This is an implementation problem, where most of the real work, and most of the value, live. It's also the exact gap Harbor Deploy, our new AI deployment offering, was built to close: the unglamorous work of connecting a chosen platform to how people work once it's live.
Continue the conversation at ILTACON
These are the conversations Harbor will be bringing to ILTACON this year. If you’re attending, join us for:
- Monday, August 24, 2:30 p.m. “Does AI Truly Break Down Knowledge and Content Barriers?” with April Brousseau, Director of Knowledge and Research Services, Harbor
- Tuesday, August 25, 10:30 a.m. “KM Roundtable 2.0: KM + AI — From Inspiration to Implementation” with Damian Jeal, Managing Director and International Lead, Harbor
Together, these sessions dig into what it takes to make AI part of everyday legal work, improving access to knowledge while maintaining the governance and controls law firms depend on.
- AI
- Knowledge management
- Tech adoption
- Show all 7
