guideCM-011
Who owns the company brain?
Nobody has settled it. The contracts allocate inputs and outputs, not the derived memory record; portability is disputed; deletion has a measured residue.
Context management for AI systems
AI Context Management
Make AI systems trustworthy by giving them exactly the right context.
Worked with:
Writing
noteCM-009
Context, memory, and state: three words for three different things
Context is what the model sees on one call. Memory is what survives the call. State is what the agent is in the middle of.
noteCM-010
What LoCoMo actually measures
Ten synthetic two-person chat logs and 1,986 questions about them. A LoCoMo score is not a measurement of memory in general.
noteCM-008
What the agent-memory vendors actually differ on
Four architectures read from their own documentation, and a benchmark dispute where one vendor's 84% was recalculated to 58.44% by a competitor's CTO.
noteCM-007
How agent memory gets written: hot path versus background
Either the agent writes during the turn and the user waits, or it writes afterwards and the memory is not there for the next question.
guideCM-006
The three types of agent memory: semantic, episodic, procedural
Facts, events, and how-to. The split is borrowed from human-memory research, and two public sources already disagree about what belongs in the third category.
New pieces by email
Want the agent reading your mail to know how to better manage its context? Subscribe.
About
Ori is a founder, keynote speaker, and AI consultant. With more than a decade of experience working with data — from data science to agentic systems and the infrastructure they run on — he deeply understands how data behaves in the wild, and how to harness it to build intelligent, scalable solutions.