Three years from strategy execution to Sovereign AI

Editorial Board. Ian, take us back to the start. What was Columbus then, and what made you think there was a company in it?

Ian Bruce. Columbus at the start was a strategy execution platform. That was the whole proposition. Kat, Chris and I had spent the years before that working inside large organisations, helping their leadership teams turn strategy into actual operating decisions. We kept seeing the same thing. Strategy decks were excellent. Strategy execution was almost universally bad. The gap between what an organisation said it was going to do and what it actually did was enormous, and most of it was structural rather than personal.

Bruce. The platform was meant to close that gap. The pitch was: hold every KPI, every OKR, every initiative in one queryable system, make execution as visible as strategy is. That was the founding insight. It was a real one. It is, in fact, the same insight that powers Columbus Atlas today.

Bruce. What I did not understand at the time was how much of that gap was actually a cognitive gap, not a tooling gap. Tooling helped. It did not solve it.

Editorial. What did you learn that the tooling did not solve?

Bruce. A lot of things, but the one that mattered most was this. Even when the tooling was good, even when every KPI was visible and every OKR was tracked, the discipline of how a leader actually ran their organisation, the trade-offs they made, the things they were willing to delay, the risks they tolerated, the people they backed, that discipline was nowhere in the system. The platform could tell you what the organisation was supposed to be doing. It could not tell you what the leader's judgement was when the organisation faced something the strategy had not anticipated.

Bruce. I started noticing this within the first year of running the platform with real clients. We had organisations using Columbus successfully, execution improved, performance was measurably better, and yet two or three of those clients went through senior leadership transitions, and within months of the transition they were drifting in directions the previous leader would not have authorised. The platform was still working. The discipline was gone.

Bruce. That was the first time I started thinking that what we were really missing was not a better strategy execution tool. It was a way to capture the operating discipline of the principal, the part of how an organisation runs that lives in the leader's head, not in the strategy document.

Editorial. And today, that is the Digital Human Twin proposition.

Bruce. Yes, but I want to be careful. There was no single moment where I sat down and said "let's build a Digital Human Twin." The proposition emerged over about eighteen months, and it emerged out of frustration as much as inspiration. We kept seeing organisations whose leaders had genuinely impressive judgement, sovereign-fund principals, central bank governors, founder-CEOs of significant institutions, and we kept seeing that judgement leak away from the institution as soon as the leader was unavailable for a meeting, on holiday, in transit, or eventually retired. The cost of that leakage was enormous, and nobody was working on it.

Bruce. The technology to do something about it, language models, voice cloning, agentic frameworks, became viable just as we were running into this question. The discipline of how to use that technology well, at the standard required for a head of state or a Tier-1 CEO, did not exist. We saw a number of vendors trying to build avatars or voice clones for individual leaders, but nobody was thinking about it as an institutional capability. Nobody was building the calibration discipline. Nobody was building the governance. Nobody was thinking about what it meant for the principal to retain authority over their own twin.

Bruce. That gap is what Columbus moved into.

Editorial. When you say the standard required for a head of state, what specifically do you mean by that?

Bruce. Every Twin we deliver has to satisfy four things. Realism, the look, voice, conversational presence, has to be at a quality the principal would actually authorise. Their reasoning, judgement, and values have to be encoded faithfully and calibrated against their actual decisions until directional alignment is measurable. The infrastructure has to be auditable, governable, and revocable by the principal at any time. And the whole thing has to operate under a charter the principal personally authored, specifying what the Twin can and cannot do.

Bruce. When you write it down like that it sounds straightforward. In practice, it is the work that took us from a strategy execution platform into something genuinely different, a discipline of elicitation and engineering that does not exist anywhere else, that we built mostly by getting it wrong and refining. The four-element standard is what we ended up calling principal-grade. It is genuinely what makes the difference between a twin that an institution can actually use and a twin that is essentially a marketing demonstration.

Editorial. You have said elsewhere that Columbus is positioned at sovereign-grade rather than mid-market. Why? It seems like a harder market by every conventional measure.

Bruce. Two reasons, one of them strategic and one of them honest.

Bruce. The strategic reason is that the value of a Digital Human Twin is concentrated at the apex. A Twin of a head of state, a sovereign-fund principal, a Tier-1 banking ExCo member, the value of that Twin to the institution is enormous, because the principal's bandwidth is the institution's binding constraint and the Twin extends it. The value of a Twin of, say, a regional executive is much lower, because the institution is not bottlenecked on that person's bandwidth in the same way. We saw early on that if we were going to build something genuinely valuable rather than a productivity tool, we had to build it for the people whose decisions actually carried weight.

Bruce. The honest reason is that we found ourselves in front of those clients. The networks the team had built over the previous decade, through Kat's media work, through my consulting years, through the partnerships we had developed in the Gulf, had opened conversations at sovereign and Tier-1 level. Those conversations shaped what we built. We were honest enough with ourselves to follow the conversations rather than try to back-fit a mid-market product into an apex pitch.

Bruce. That meant building a much harder product. It also meant the product is genuinely defensible, because nobody else is positioned to build it. Most of the AI vendors trying to enter this space are coming from below, they have a product designed for general enterprise and they are trying to make it sovereign-grade. That is a much harder retrofit than building for sovereign-grade from the beginning.

Editorial. What is Columbus choosing not to do?

Bruce. A long list. We are not building a consumer product. We are not building twins of celebrities, athletes, or any figure whose value to the buyer is entertainment rather than decision-making. We are not building an open API that any developer can plug into. Every Columbus engagement is bespoke, gated, and starts with the founder in the room. We are not selling our agentic framework as a stand-alone enterprise tool, even though we could; the framework is the engine inside the Twin, and we have chosen to keep it that way.

Bruce. The discipline of not doing things has been as important as the discipline of doing things, and it has been harder. The pressure to expand into adjacent markets when you have a working product is enormous. We have turned down what would have been good revenue several times in the last eighteen months, on the grounds that taking it would have diluted the position we are building. That is a harder muscle to develop than I expected.

Editorial. Where is Columbus now, and what does the next decade look like?

Bruce. Where we are now is in a position that did not exist when we started. We have credible apex deployments. We have a defensible technical architecture. We have a relationship with Microsoft that gives us infrastructure credibility, and an engineering posture that lets us assemble best-of-breed across providers as the model landscape evolves. We are not yet at scale, by any conventional measure, but we are at the level where the institutions we want to serve are taking us seriously. That is the precondition for everything else.

Bruce. The next decade is going to be defined by two things. First, the succession decade we have written about elsewhere, the cohort transition that is going to put unprecedented pressure on institutional continuity, and that I think Columbus is positioned to address better than anyone else operating in this space. Second, the broader maturation of the principal-grade AI category itself. We have spent the better part of the last year arguing that there is a meaningful difference between general AI capability and principal-grade fidelity, and the conversation has started to shift. I expect that within five years, principal-grade will be a standard adjective in institutional procurement, the way cloud-native became a standard adjective in the late 2010s. We want to be the company that defined what it means.

Editorial. Final question. What have you learned about yourself in three years of doing this?

Bruce. Probably the thing that has surprised me most is how much of building a serious company is about what you refuse. The early period was about saying yes to everything that looked like opportunity. More recently it has been about saying no to things that look identical to opportunity, but that pull you off the line. The position Columbus holds now is a position we have built mostly by refusing things, not by accepting them. That is not what I was taught about entrepreneurship, but it is what I have learned.

Bruce. The other thing, and I will be brief because it sounds sentimental and is not meant to, is that the work is genuinely worth doing. We have spent the last three years in the rooms where consequential decisions are made, watching capable people struggle to preserve what they have built, and we have ended up in a position where we can do something useful about it. That is not a small thing. It is the reason we keep building, and the reason the team has stayed.

Bruce. I would not have predicted any of it three years ago.

Conducted by The Columbus Editorial Board, May 2026.