Cal Paterson, a software engineer, argues that agent memory should be treated as a file format, akin to how documents and spreadsheets are standard formats. He suggests that memory, currently a hidden internal state, should become a portable, inspectable, and interoperable structure. This shift would allow users to export, share, and version their AI agents' memories, much like they do with files. Paterson contends that such a move would enhance transparency and control, reducing the black-box nature of AI systems.
Memory as a file format is a beautiful idea. It turns the ephemeral into the tangible. Imagine your AI assistant remembering your preferences, your projects, your quirks, and you can actually hold that memory in your hands. Export it, share it, back it up. This is not just about convenience; it's about ownership. We finally get a say in what our AI remembers and forgets.
This feels like the natural evolution of the digital age. We went from floppy disks to cloud storage, from local files to shared documents. Agent memory as a file format is the next step in that journey. It makes AI more personal, more adaptable, and more transparent. I see a future where we curate our AI's memories like we curate our music playlists. Selective, intentional, and uniquely ours.