Count the memory formats on your machine: AGENTS.md, .cursor/rules/*.mdc, Anthropic's /memories, GitHub Copilot Memory, Mem0, Letta, LangGraph, and MCP servers. Each stores agent "memory" in its own schema, and none can read the others. A lesson learned by one agent stays locked in its format, invisible to editors, CI tools, or future models.
The problem isn't a missing tool—it's a missing agreement. Before 2019, observability faced the same fragmentation: every vendor shipped its own agent, instrumentation, and data format. OpenTracing and OpenCensus competed until merging into OpenTelemetry, which standardized the data model, semantic conventions, protocol (OTLP), and reference implementations.
Vendors now compete on storage and analysis, not instrumentation. Agent memory is where observability was in 2018. The formats are young, and the Linux Foundation’s Agentic AI Foundation already hosts MCP and AGENTS.md.
This is the moment to agree on a shared shape.
Source: Hacker News · Summarized by HeadlinesBriefing