Published
Why model changes are system changes
A longer reflection on what happened when I swapped models inside the same OpenClaw workflow and realized the model was doing much more than generating prose.
Read noteNotes
A few longer notes from building with these tools.
This is where I am turning some of the more durable project lessons into fuller writeups.
In progress
Why long-lived agent state needs progressive disclosure
A note on what happens when a useful agent workflow accumulates too much state, and why a smaller index plus deeper context on demand works better than one giant memory file.
In progress
What layered evals are actually buying you
A broader reflection on why a single quality check is rarely enough once a system is summarizing, recommending, or acting on behalf of a user.
In progress
Why a skip button can mean three different things
A note on Desk Flex, feedback semantics, and why one simple user action can mean very different things to the person and to the system.