Writing ( 02 )
Research-backed notes on how AI systems remember, decide, act, and earn trust.
Evidence before opinion.
-
Mar 7, 20267 min read · 4 sourcesProduct design · Agentic UX
Confidence is the missing layer in AI interfaces
Why model confidence is not enough—and how evidence, consequence, and reversibility should decide when an agent acts, proposes, asks, or stops.
Read essay -
Feb 28, 20268 min read · 4 sourcesAgentic UX · Memory design
How to approach UX for agentic systems
Long context is not reliable memory. A practical model for selective, correctable context backed by four long-context studies.
Read essay
Substack
Thoughts and work in progress also live on X.