SecondSourceJudgment rebuilt from primary sources

Looking back · SecondSource reading of July 14

AI agents' memory currently sits outside the model: stored as text and placed in the input window each time, not written directly into the model's weights, the model's own parameters.

Concept illustration: a person with dark hair and dark clothing, seen from behind at the right edge, holds up a large white board with both hands, and a black band lettered SECONDSOURCE runs across its top. On the board, a box lettered MEMORY on the left holds a row of upright red-orange cards marked with white lines, and to its right an open laptop in front of a pale pink rectangle shows a speech bubble with grey lines above a red-orange block with white lines; the box and laptop sit side by side on the same thin line. Printed on the image: Per the July 14 reading, agent memory is text outside the model, put in the input window each time.

On this reading, the work on memory layers that enterprises build for agents over the next year centers on how to store, retrieve and place memory in the window. The reason is that whether new knowledge written into weights can be kept long term, without erasing old abilities, hasn't been proven. The test SecondSource set: if by July 14, 2027 two or more frontier labs ship memory written into weights in production products, this reading is wrong (July 14 original reading). On November 5, Trajectory, a startup building continual-learning products, appeared on Sequoia's podcast (Sequoia, 2026-11-05). Trajectory is a startup, not a frontier lab, and doesn't count as a counterexample under this test. How we checked →

Source and checks

This item ran in the brief of Nov 11, 2026, alongside 15 other items. How we checked it, the quote checks and what was still missing that day are on that issue's How we checked page.

Cite this item

SecondSource (2026). "AI agents' memory currently sits outside the model: stored as text and placed in the input window each time, not written directly into the model's weights, the model's own parameters.." https://secondsource.io/en/news/2026-11-11-7/. Accessed access date.