[NOW]
Last updated: April 2026 — Warsaw / Austin
This is a /now page. A snapshot of what's occupying my attention at this moment in time.
Deep in the architecture phase. Working on the Hilbert curve spatial indexing and temporal decay functions. The core question I'm trying to answer: how should an AI agent's memory of a place change as that place changes over time?
Live at getarchflow.io. Currently focused on improving the consultation-to-render pipeline and onboarding more interior designers. The 72-hour delivery promise is holding up well.
Exploring the SaaS layer on top of the open-source core. The question: what does a commercial privacy-preserving AI service look like when the underlying architecture is zero-knowledge by design?
Context as infrastructure. We've spent a decade building compute infrastructure for AI. The next decade will be about context infrastructure — the systems that determine what an AI knows, when it knows it, and how certain it should be.
The optimization trap. When every agent optimizes for the same signal, the signal degrades. I wrote about this recently — the Nash equilibrium problem in AI-driven systems. What happens when everyone's AI is optimizing against everyone else's AI?
Spatial memory for embodied AI. Current vector databases are essentially memoryless about space and time. LOCI is my attempt to fix this. But the deeper question is: what does it mean for an AI to have a sense of place?
Thinking in Systems
Donella Meadows
The Alignment Problem
Brian Christian
Papers on CRDT synchronization
Various
Split between Warsaw, Poland and Austin, Texas. The time zone gap is a feature, not a bug — it forces async-first thinking, which turns out to be excellent training for building distributed systems.
Inspired by nownownow.com. Updated roughly monthly or when something significant shifts.