Agent Memory: Where Does Knowledge Live? Webinar! (Ep 2)

​If you ask a team how their agent's memory works, you'll almost always hear a database name back, because that's the decision that felt consequential at the time and it may be the least consequential choice on the table.

​In episode two we go deep on the first of the four modules from Are We Ready For An Agent-Native Memory System? (Zhou et al., SJTU/Tsinghua/MemTensor), the layer that sets a retention ceiling nothing downstream can lift.

​Storing raw verbatim turns beat every summarized and compressed alternative tested, deepening a hierarchy moved the numbers barely at all, and swapping in a graph store beneath an unchanged extraction pipeline did nothing, meanwhile the most structured systems ran roughly forty times slower for well under twice the utility.

​We'll work through where abstraction costs more than it saves, what structure is actually good for, and which side of that trade your workload sits on.