AI summaries are table stakes. Persistent speaker identity is the moat.

Summaries are not the hard part

Every meeting product now wants to lead with AI summaries. That is useful, but it is not the deepest part of the problem. A summary is a compressed view of one meeting. Identity is what makes the archive cumulative.

If you are building for agents, the question is not only "what happened?" It is "who said it, and does that person stay the same across the rest of the corpus?"

Why identity is the real unlock

Without stable identity, a meeting archive resets every time. You can read the current transcript, but you cannot easily ask what the same person said last month, or trace how a position changed over several discussions.

That makes the archive much less useful for humans and almost useless for agents that are trying to understand a work history rather than a single conversation.

How Transcripted handles it

Transcripted keeps each meeting local and structured. The human layer is a Markdown transcript. The machine layer is YAML frontmatter. The corpus layer is the capture folders.

Persistent speaker identity sits underneath that structure and helps the archive stay coherent as it grows. The same person can carry through multiple meetings, so the archive does not have to start from scratch every time.

Why this matters for agents

An agent is only as good as the context it can retrieve. If it does not know that the same voice appears in multiple meetings, it cannot answer cross-meeting questions with confidence.

That is why the rest of the product matters too. The starter prompt, MCP server, and CLI all point at a corpus that needs consistent structure and identity if the agent is going to use it well.

The broader implication

Speaker identity is not a fancy extras feature. It is a data model decision. Once the archive knows who is who, the product starts to feel less like a recorder and more like a private meeting memory system.

That is a much better fit for the next wave of personal agents.