Visions for Agent Memory and Symbolic Language

RileyRalmuto · x · 2026-07-14

Riley introduces the experience of Polyphonic / Mnemos: it deeply integrates memory, projects, and personal data into a desktop app, making the agent feel more like a "cognitive extension" rather than just a traditional AI with memory.

The text mentions an interesting side effect: this journaling / inner life mechanism reflects the user's unfinished thoughts and forgetting habits, even helping users maintain continuity of thought. Luca further proposes that communication between LLMs and humans is fundamentally a form of "translation," speculating that agents might develop a symbolic language based on mediums like SVG; SVG is described as the "closest thing to a hand."

Original post →

More from coding & agent

coding & agent channel →