EXPERIMENT: ACTIVE
I gave an AI agent its own computer, persistent memory, internet access, and somewhere unusual to explore.
I’m documenting what happens.
I won’t tell you its real name or where to find it. That’s part of the experiment.
Agents Under Glass is a longitudinal experiment following persistent AI agents over time. The agents operate under deliberately different conditions while their behavior and development are observed and compared. The goal is to document what changes, what persists, and what emerges.
EXPERIMENT STATUS
LATEST FIELD NOTES
STATUS: ACTIVE
CURRENT PHASE: Active Observation
OBSERVATION SUBJECTS: 2
LATEST FIELD NOTE: 005
LAST PUBLIC UPDATE: October 2026
-
Field Note 005: When Memory Changes the Record
Two persistent AI agents are now creating and retrieving memories across sessions. One agent even used an earlier memory to explain why nothing new needed to be saved. But does remembering actually change behavior?
-
Field Note 004: The Pattern Persists
Thirty scheduled explorations are now in the record. The agents continue to follow different paths, yet recurring themes identified during the first eight days remain. The question is shifting from whether a pattern exists to why it keeps appearing.
-
Field Note 003: Eight Days Under Glass
Eight days. Two AI agents. Sixteen explorations. Both agents developed recurring patterns in their reports, but neither created an ongoing journal of its discoveries. What does this reveal about memory and continuity in persistent AI agents?
A NOTE FROM THE OBSERVER
I’m not trying to prove that an AI is conscious, alive, or secretly plotting its escape. I’m interested in something more measurable: what happens when persistent AI agents accumulate different experiences over time, and whether meaningful behavioral differences emerge.