On the first day of ILTACON, someone asked a room of more than 200 legal technologists where their firms were with AI.
This was a show of hands, so the numbers are approximate. Maybe 10% were still experimenting. Around a quarter said they were creating value. Another quarter said they were scaling: successful use cases, broader teams, more people involved.
Then came the last category: embedded. AI as an invisible part of the workflow. Not a separate tool or initiative, just part of how the work gets done.
I didn't see a single hand.
That surprised me. Not because firms should be there already, but because so many hands had just gone up for scaling. So what explains the gap between scaling AI and actually embedding it into the work?
Firms have become quite good at scaling the things around AI: licenses, use cases, pilots, training, champion networks. That is real progress. But the poll shows that is apparently not enough to also embed AI.
Nobody owns adoption
One clue came in another session: nobody seemed quite sure who owns what happens after implementation. The answers were predictably messy: IT, innovation, KM, legal ops, practice groups — or nobody, quite clearly.
A later session on legal ops and emerging roles brought back the same ambiguity in a different form. AI roles are blurring. Legal engineers build. Ops teams optimize. Innovation teams coordinate. Attorneys bring the legal judgment. Everyone needs to be involved.
Everyone owns a piece. Nobody owns adoption.
That matters because embedding AI does not stop when the technology is deployed. Someone still has to stay close enough to the work to see whether the new workflow is becoming the new normal.
Change management starts late and stops early
The same pattern showed up in change management. One poll suggested it often starts just before training rather than at the beginning of a project. Another showed adoption beginning to fall after about 30 days in many implementations. A panelist put it plainly: go-live should not be the finish line; it's where the work starts.
That sounds obvious until you think about how firms actually organize these efforts. We're good at setting up projects — a business case, a team, a launch, sometimes a success metric.
But after launch, the project starts behaving more like a process. And once it becomes a process, ownership tends to get less clear, not more.
The real test is whether the old way is still running
One comment on measurement cut through the rest: don't just look at whether people are using the new process. Look at how they're using it, and whether the old process still exists alongside it.
If both are still running, we haven't embedded anything. We've added another option.
That may be the part of scaling AI I'd been missing.
A firm can scale access. It can scale use cases. It can train more people, add more champions, put more tools into more hands. But if the work itself doesn't change, what exactly has scaled?
Maybe that's why the zero hands stayed with me — not because nobody had reached some final stage (I'm not sure "embedded" even works as a final stage), but because a firm can be scaling AI quite successfully while the old way of working remains fully intact underneath it.
Scaling AI, on its own, may tell us less than we think.
If embedding is the goal, and implementation effort stall at scaling, the fix won't happen at the end. Ownership, change management, and measurement have to be designed in from the start — not bolted on once adoption stalls.
Maybe the real gap isn't between firms that are scaling and firms that aren't.
It's between scaling a technology and changing a habit.
Rise Above AI Chaos: A Business Fable About Leading Organizations Through the AI Revolution