Women, AI, and the rooms where it is being decided

Editorial Board. Kat, we want to ask you about women at the forefront of AI, but we want to do so without writing the piece every other publication has already written. Where do you actually think the conversation should be?

Kat Gref. I appreciate that, because the standard piece is exhausting. Most of what gets written under the heading women in AI is either a complaint or a list. The complaint is that there are not enough of us, which is true but has been true for ten years and has not been resolved by being written about. The list names the same handful of people repeatedly, which has its own purpose but is not actually a conversation.

Gref. What I find more interesting, and what I think the real story is, is that the rooms where AI is being decided are different rooms than the ones the commentary assumes. The published commentary mostly looks at frontier model labs in San Francisco and London. The actual decisions about how AI gets deployed at consequential institutions are being taken in different places, in Abu Dhabi, in Singapore, in Brussels, in Riyadh, in central bank governance committees, in sovereign-fund investment offices, in Tier-1 banking risk committees. The composition of those rooms is, in places, more interesting than the commentary suggests.

Gref. The commentary has been describing one kind of room and missing another. The rooms I am actually in look different from the rooms the commentary assumes.

Editorial. Say more about that. The Gulf is sometimes characterised, in Western media, as the opposite of mixed.

Gref. I know. And I want to be careful here, because I do not want to overstate the case in either direction. The Gulf has its own complicated history with women in senior roles, and I am not going to pretend that the picture is uniformly good. There are places where it is, and places where it is not.

Gref. But the picture in the AI procurement conversations specifically, at the level of institutional decision-making, not at the level of frontier research, is more interesting than the casual reader would expect. I have sat across from women running technology procurement at sovereign funds. I have presented to women running risk committees at Tier-1 regional banks. I have been in cabinet-adjacent meetings where the senior policy adviser on AI governance was a woman, and the chief of staff was a woman, and the most pointed questions came from a woman who had been in the role for fifteen years and had read every paper that the Western commentary considered relevant.

Gref. The contrast that surprises me is not Gulf-versus-California, because I do not have direct experience of California to draw on. The contrast is between the commentary I read and the rooms I am actually in. The published narrative about institutional AI assumes a particular kind of room, one that is mostly male, mostly Western, mostly research-trained. The rooms I am in look different. They are more diverse by background, more institutional in posture, more concerned with governance than with capability headlines. And those are the rooms making the deployment decisions that actually shape how AI gets used at scale.

Gref. So I am not making a claim about the relative composition of laboratories I have not been to. I am making a claim about the gap between the commentary and the institutional reality. And that gap is real.

Editorial. You are saying the centre of gravity for institutional AI is shifting toward those rooms?

Gref. I am saying it has already shifted, and the commentary has not caught up. The frontier model conversation, what scale, what training data, what alignment technique, is dominated by a small number of laboratories, and that conversation will continue to matter. But the deployment conversation, what models get used by what institutions, under what governance, to do what work, is much broader and much more global. It includes the institutions Columbus serves, and it includes a lot of women who are not visible in the frontier-lab story.

Gref. I think the most interesting question in AI right now is not who builds the models. It is who decides where they get deployed at the institutions that actually matter. And the answer to the second question is much more diverse than the answer to the first. We should be writing about that.

Editorial. What makes you say the deployment conversation is the more consequential one?

Gref. Because deployment is where the social effects of AI actually land. A frontier model is a research artefact until an institution deploys it. The choice of whether a sovereign fund uses AI in its investment decisions, whether a central bank uses AI in its supervisory work, whether a national health service uses AI in its triage protocols: these are the choices that determine what AI does to societies. The model itself, in isolation, does very little.

Gref. And those choices are being made by people whose training is in policy, in governance, in regulation, in operations, not in machine learning research. Many of those people are women. They are not on the conference circuit because they have institutional roles, not personal brands. But they are the ones writing the procurement specifications, the ones approving the audit frameworks, the ones deciding which vendors are credible. The shape of AI in the next decade will be set by them at least as much as by the model labs.

Editorial. Columbus has built infrastructure specifically for those buyers. Has the gender composition of your buyer base shaped the product?

Gref. That is a sharper question than people usually ask, and the honest answer is yes, but probably not in the ways the question expects.

Gref. The product has been shaped by the concerns of our buyer base, and those concerns have been shaped by who the buyers actually are. The concerns we hear most consistently are provenance, governance, revocation, charter authority, audit trails, calibration evidence, these are not the concerns the wider AI commentary emphasises. The wider commentary emphasises capability, productivity, and benchmark performance. Our buyers care about those too, but they care about them after they have been satisfied that the system can be governed.

Gref. I would not say this is uniquely a function of female leadership in our buyer base, but I would say that the female leaders we work with tend to be more rigorous on the governance side: more demanding about audit, more careful about charter language, more interested in revocation as a real capability rather than a marketing feature. That pattern is consistent enough that we have built our product around it.

Gref. The result is, I think, a stronger product. Governance is not a soft requirement. It is a hard one, and it will become a harder one as the regulatory environment tightens. We have built for the standard our buyers actually require, and our buyers have, in a non-trivial proportion, been women operating at very senior institutional levels. Those things are connected.

Editorial. You have used the phrase performing inclusion versus being present before. Will you say more about that?

Gref. It is a frustration of mine, yes. Most companies in the AI space have a public-facing story about women in their organisation, and that story is mostly performance. The marketing materials show the diverse team. The investor decks include the diversity slide. The CEO speaks at events about how important inclusion is. Underneath, the actual presence of women in the rooms where decisions get taken is often much thinner than the public-facing story suggests.

Gref. I am not going to claim Columbus has fully solved this; we are still a small team and the composition will shift as we grow. But the difference I try to insist on is that being present is not a function of marketing. It is a function of who actually decides things. If you are in the room when the architecture is decided, when the pricing is set, when the engagements are accepted or refused, when the governance is written, you are present. If you are mentioned on a slide but not in the room, you are not.

Gref. The test of any company’s record on this is not what they say about it. It is whether the women they employ are visibly making decisions that the company then acts on. That is a much harder test, and it is the only one that matters.

Editorial. What advice would you give to women who are entering AI now, particularly in regions where the path is genuinely harder?

Gref. I am cautious about giving advice, because the situations vary so widely that any single piece of advice is wrong somewhere. But there is one thing I would say.

Gref. Most of the public path into AI right now goes through the frontier research narrative: get a PhD, work at a model lab, build a reputation as a researcher, eventually move into governance or policy. That path is real, and for the right person it is excellent. But it is not the only path, and in some ways it is no longer the most important one. The deployment path, the path that goes through institutional roles in policy, governance, regulation, procurement, and applied research, is shorter, more directly consequential, and more open. The institutions that matter for AI’s actual social effects are recruiting heavily for that path right now, and they are recruiting in a much more open way than the frontier labs are.

Gref. If I were entering the field now and I cared about shaping AI’s effects rather than building the underlying technology, I would seriously consider whether the deployment path is the more leveraged one. For many people, in many places, it is. And it is the path that has, in my experience, more women already on it.

Editorial. Final question. What has surprised you most about being a co-founder of a company at this level?

Gref. That so much of it is about endurance. The story of building a company is often told as a story of inspiration and bold choices. There are bold choices, but they are perhaps five per cent of the actual time. The rest is endurance: turning up, holding the line on the standard, refusing the things that pull you off course, doing the unglamorous work of governance and discipline week after week.

Gref. I think that is also, in a quiet way, why the women I work with at sovereign and Tier-1 level are good at what they do. The path most of them have taken to get there has been one of sustained endurance over decades, in environments that were not always welcoming. They are not surprised by the discipline the work requires. They have been doing it for a long time.

Gref. It is, I think, an underrated qualification. And it has been underrated long enough that the people who have it are now in some of the most consequential rooms in the world. The commentary has not noticed yet. It will.

Conducted by The Columbus Editorial Board, May 2026.