The room in Alicante was noticeably warmer on AI than it was twelve months ago. A year ago, the questions were mostly defensive: Does this actually work? Will it hallucinate on me? What happens when it gets something wrong? This year, those questions were still there, but they sat in the background. The conversations that took up the most oxygen (on the main stage and in the networking breaks) were different. They were quieter, more searching, and a lot less about the technology itself.
Why the room felt different
The reason, I think, is that the GCs in the room have moved from being the audience to being users.
A year ago, when AI came up at events like this, most GCs were talking about something they had not yet touched. They had read about it, sat through a vendor demo or two, seen a colleague's experiments from a distance. The relationship to the technology was second-hand.
That has clearly changed. In almost every conversation in Alicante, the GC across the table had been using a general-purpose AI tool themselves, e.g. Claude, ChatGPT, Copilot, sometimes more than one, and often for something relatively simple – e.g. stress-testing a clause, drafting a memo, summarising a long document.
The opinions were mixed, and that is the point. They have run into the limits. They know what the general tools get wrong on legal-specific work.
But they are also not afraid of the technology in the way they were twelve months ago. Personal use has done what no demo ever could: it has made the thing concrete. They know what it can do, what it cannot, and roughly where the line sits between the two. Whatever they think about the next frontier of AI (i.e. agents), they are thinking about it from a position of having actually handled some of the underlying technology.
The so-what, I think, is that the next twelve months will see a lot more GCs building or configuring their own things. Not full legal AI platforms (those are not in the GC's job description, and shouldn't be), but things like assistants to help lawyers work slightly faster. These things are now within reach of a motivated GC and a willing IT partner, and a meaningful number of GCs are going to try them.
The implication for the rest of us (vendors, advisors, partners) is that the buying conversation is about to get more specific. A GC who has spent six months running their own internal copilot has a much sharper view of what it cannot do, and where they actually need help. The conversation in 2027 will be less "convince me this works" and more "here is what I have already tried, here is where I got stuck, what do you do that I can't do myself." That is a healthier conversation for everyone involved. It also raises the bar.
As a result of this shift, three things have changed in what GCs are spending their energy on. None of them are really about AI itself. They are about what AI is doing to the legal function: how it's trained, how it's governed, and how it relates to the firms that have been its partner for the last fifty years.
1. "We won't be able to train the next GC"
This was the worry that came up most often, and the one with the least clear answer.
The case for AI in a legal department is usually framed around speed and leverage: the first draft of an NDA is done in two minutes instead of two hours, and a junior lawyer's afternoon is freed up for something more interesting. That framing is appealing to operators. It is much less appealing to the people who are responsible for producing the next generation of senior counsel.
The concern I heard, repeatedly, was that the craft of being a good lawyer is built through the grind that AI is now compressing. You learn judgment by sitting with a hundred bad versions of a clause and figuring out, slowly and uncomfortably, why one of them is right. You learn what a difficult counterparty actually wants by being on the wrong end of the call enough times. You learn how to push back on a commercial colleague by getting it wrong, getting feedback, and trying again. The hours that AI saves are, in large part, the hours where formation happens.
GCs aren't reactionary about this. Most of the ones I spoke to are deploying AI in some form. But they don't yet have a clear answer to: If we shorten the path, what replaces the path? What does the new training arc look like? What does a five-year-qualified lawyer who grew up with AI actually know compared to one who didn't? And which gaps will only show up when that lawyer is in a senior seat and the AI can't help anymore?
This is going to be one of the defining people questions of the next few years, and I came away convinced that GCs feel personally responsible for getting it right.
2. "Where does the data actually go?"
The second worry was less philosophical and more operational, but it has a similar quality of unanswered questions sitting underneath a familiar surface.
A year ago, the data conversation with GCs was a relatively simple one. The question was "is my data safe?"
The conversation has become a lot more nuanced. "Is it safe" is no longer the right framing, because it implies a binary that does not match how these systems actually work in practice.
The questions I heard in Alicante were more granular. Is the tenant isolated? SOC 2 Type 2? What happens to a document when it passes through a model, and even if the model isn't being trained on it, what artifacts exist, for how long, and who can see them? If a regulator asks the GC tomorrow to draw a map of where a specific contract's contents have travelled, can the GC actually draw that map?
These are not unreasonable questions, and they are not paranoid questions. They are the questions of a function that knows it will be the one in the chair when the board, the regulator, or the customer asks. GCs are not looking for reassurance. They are looking for evidence. The kind that holds up when they're the ones being asked to defend it.
3. "We want firms as advisors, not drafters"
The third worry was about a relationship, not a technology. It was also the one that felt most quietly seismic.
For a long time, the unspoken bargain between in-house teams and outside counsel was that the firm did the work the in-house team couldn't or wouldn't do itself: the drafting, the late-night turns, the volume. The relationship was sized to that bargain. Panel arrangements, billing models, even who you put on which file, most of it assumes that drafting and turning documents is the substance of what firms provide.
What I heard at the summit is that this assumption is breaking. As "work" moves in-house, accelerated by AI agents that can take a first or second pass at the kind of work that used to fill associate timesheets, GCs are starting to ask a sharper question of their firms: What is the part I actually need you for?
The answer most GCs landed on was some version of judgment – e.g. the hard call on a novel question, the view of how a regulator is thinking this quarter, or simply their instinct, which has been built over decades. That is what they want their outside counsel relationship to be about. Not the draft.
That sounds like a small reframing, but it changes how matters get scoped, how firms get paid, and what kind of partner gets promoted inside the firm itself. Several GCs told me, almost in passing, that they were quietly rethinking their panels with this in mind.
What it adds up to
The headline from Alicante, if there is one, is that the conversation about AI in legal has moved on. The defensive questions of last year (does it work, can I trust it, will it embarrass me) haven't been resolved so much as absorbed. The questions GCs are spending their energy on now are about second-order effects: what AI is doing to the people they're responsible for training, the data they're responsible for governing, and the partners they're responsible for choosing.
These are harder questions, and they don't have vendor-shaped answers. But they are the right questions, and the fact that the room is asking them, out loud, in detail, with each other, is the most encouraging thing I took home.
