The longer I work on complex problems, the more I grow skeptical of pure reasoning.
It’s too neat — too good at defending its own blind spots. Not humble enough to its low survivability odds in the real world. Even the simplest organism that survived evolution is complex. Most products we buy today are extremely complex - a person famously tried to create a simple toaster from scratch and realized the enormous supply chain complexity behind it.

So in my work, I use reason to build the container, then let chaos and complexity fill it.
I push ideas in every direction — what builders feel, what buyers aspire to, what the world might actually want.
Large language models help stretch that chaos even further, surfacing edges I’d never see alone.

That’s how the biosecure-shelter project began: one sprawling storm of “what-ifs” — from inflatable school domes in polluted Beijing to Antarctic research pods and high-design bunkers — before a coherent, minimal form emerged. And the direction it ended taking was totally unexpected to all. It was impossible to reason about what would work. The world is often too complex to reason about.

And it’s the same pattern I’ve noticed in work tests with EA orgs.
The short answers with concrete actions do the worst. The unnervingly long, sprawling reasoning chains thinking of this and that win. But, and this is an important "but": At the end one should show how all that complexity results in simplicity in the real world. A single line in an email. Three lines in a Google Sheet.
The answers optimised for tidy reasoning don’t make it.

So after chaos comes "package compression": The output encapsulates a lot of that complexity, but it is short. Or the product does not have many moving parts, at least not on the outside.

And in any case, most of it fails. That’s the point.

Every quick failure collapses uncertainty faster than a week of careful thought.

Expand → distil → test → learn → repeat.

It’s not anti-reason — it’s reason with the humility to step aside when reality starts teaching.

I apply this approach in my own work designing and running complex operational systems — both within my own organisations and for clients. If this resonates, I currently have limited availability for collaborations or discussions on how to make messy ideas work in the real world.

Author works on operational design and resilience projects (independent and for clients). Limited availability for collaborations. Views personal.