How do you build a whole country's AI strategy? Not a slide deck about one — the real thing: months of work, hundreds of source items, and a team of economists, lawyers, and developers who all work differently.
That's the project Jaspur Højgaard took on, consulting the Faroese government on a national AI strategy. In his Usable Fragments talk, he shares what he learned using Usable as the knowledge base underneath it all.
Not a technical project
The first decision was to treat this as bigger than tech. A national strategy touches everyone — and some of the most exposed people aren't engineers at all. Teachers, for instance, have spent years building curriculums and exams that generative AI now challenges head-on. So the team went out and gathered many voices: interactive workshops across the public sector, each pointing in a different direction, all of it needing to be captured, synthesized, and turned into something actionable.
That's a brutal knowledge-management problem. Here's what made Usable fit.
Available wherever he worked
Over the life of the project, the leading AI tool changed roughly ten times — Usable Chat, Cursor, Claude Cowork, and whatever came next. None of those switches cost him anything, because the knowledge didn't live in the tool. It lived in Usable, reachable over MCP. Switch the tool, plug back in, and everything's still there. No mandate that the whole team use one specific app.
Available to everyone on the project
A national strategy isn't a one-person job, and the knowledge base couldn't be either. One workspace, many contributors. There's no "which file is the latest version" problem, because the fragments are the source. And when new people join — which happened as the team grew from a small technical core to economists and lawyers — they ramp up by browsing and asking, not by being onboarded from scratch.
Synthesis and analysis on top
Once everything was captured, the analysis got fast and concrete. The AI drafts from real fragments instead of generic prompts: pulling relevant sources for a first cut, running cross-cutting analysis to compare themes across every workshop, and — the part Jaspur values most — surfacing what's missing on purpose. The hardest answers aren't the wrong ones; they're the incomplete ones.
"It's not that what it provides is incorrect. It's just not complete."
That's where humans stay essential, and where having all the source material in one place pays off.
Memory as a deliverable
Which leads to Jaspur's closing idea, and the one most worth sitting with: the project's memory can be a deliverable — not a footnote to it. A national strategy doesn't have to be only a finished PDF. It can be a living memory you can ask: what did this workshop conclude, why was this decision made, what did we hear from Estonia? You can still produce the beautiful report. But you can also hand over the workspace.
Watch the full talk
Jaspur goes deeper on briefing the AI before every meeting, picking the right model per task (including Faroese-language work), and a candid Q&A on government data safety and permissions. It's the third of four talks from Usable Fragments 2026.
Usable is knowledge management for AI agents — the shared memory layer behind the work → usable.dev
Watch the full series → Usable Fragments 2026 on YouTube