How One Developer Orchestrates Three AI Agent Projects Using Shared Memory Infrastructure

A developer running three parallel projects — Bloomii, Kalceo, and Ekioo — built a shared infrastructure on a kanban tool called KittyClaw to prevent AI agents from restarting without context on every run. Each project uses a structured directory containing a shared context file, automation pipelines, and per-agent memory files that persist lessons learned across sessions. A counter system tracks how often each lesson recurs, promoting frequently reapplied rules from a dynamic memory file into a stable skills file once they hit a threshold. An automation pipeline ensures agents only pick up new tasks when their current one is complete, preventing uncontrolled parallel execution. The core argument is that consistent cadence — not sheer volume of agent runs — is what allows AI-assisted projects to compound knowledge over time.