The Next Decade: What I Intend to Build
A plain statement of direction — the problems that justify a decade of attention, and the standard I would like this work measured against.
The selection criterion
There is no shortage of important problems, so the question is which ones justify a decade rather than which ones matter.
My filter has three parts. The problem must be one where the constraint is execution rather than knowledge, because problems awaiting a scientific breakthrough are not ones I can usefully accelerate. It must be one where a working example changes what others believe is possible, so that success propagates beyond the specific instance. And it must be one where I have some genuine comparative advantage, rather than merely a strong opinion.
Three commitments
Applying that filter produces a short list.
The first is sovereign AI capability for states that will not build frontier models — which is most states. The useful work is not another national model but a replicable pattern for controlling data, adaptation, evaluation and deployment. The knowledge exists in fragments; the execution is the constraint.
The second is identity and payment infrastructure as the substrate for inclusion. This is well understood in principle and unevenly implemented in practice, and the gap between the two is almost entirely institutional.
The third is the governance layer that makes both durable — the unglamorous work of delivery capacity, procurement reform and institutional continuity, without which the first two produce impressive pilots and little else.
How I would like it judged
I would prefer this work to be assessed against three questions rather than against activity.
Did anything continue to operate after I stopped paying attention to it? That is the real test of whether an institution was built or merely a project run.
Did the people it was meant to serve actually use it, at scale, over time? Coverage figures and launch attendance are not evidence of this.
Did the capability transfer? If the systems depend on the same small group of people a decade in, the work has not succeeded — it has merely persisted.
These are harder standards than the ones this field usually applies, and they are the ones I would want applied here.
A note on timelines
A decade is a long time to commit to a direction, and the honest position is that some of this will be wrong. The specific technologies will change and some of the assumptions behind these commitments will not survive.
What I do not expect to change is the underlying shape of the problem: capability without institutions is fragile, institutions without capability are inert, and the work of joining them is slow, unglamorous and worth doing. That is a stable enough foundation to plan on.