Decision-OS V11 Note: Forget for Evolution: Reconnectable Forgetting for Agentic AI

This note introduces Reconnectable Forgetting as a minimal memory principle for agentic AI. It addresses a current operational pain in long-running AI agents: as agents operate over longer histories—conversations, files, tool calls, code changes, execution traces, and human feedback—token cost and context load can grow faster than useful intelligence. Decision-OS V11 reframes forgetting not as loss, but as controlled compression. Raw histories are compressed into reusable structures, anchored by evidence and hashes, and indexed by time so that future intelligence can re-evaluate the original context when needed. The goal is not to remember everything, but to forget safely enough that future intelligence can reconnect. In the Decision-OS lineage, V11 follows V10 Survival-Bounded Planning. V10 asks how an agent survives without breaking its aspiration. V11 asks what the surviving agent must stop carrying in full in order to continue evolving.