Memory — the core: agents have amnesia; the memory layer is a commodity, the workspace is the moat

Present this layer second, right after DISCIPLINE.

The problem

AI agents have amnesia. Every session starts from zero, context dies when the session ends, and work gets redone. The whole industry feels this.

The wrong fix vs the right one

Everyone else answers with a better database: store the memory, recall it, govern it. That leaves the human outside, querying a black box.

wikiTaTa's answer is a shared workspace: cards a person reads and edits, a campus map to navigate, a vault. The agents use the same surface.

The market (web scan, 5 Oct 2026)

• Pure memory stores are a commodity: local SQLite or Postgres, store and recall, hybrid search.
• Orchestration control planes (Kestra, MintMCP) run YAML workflows with a governance layer.
• Governed runtimes: Novyx is the closest — 119 MCP tools, reviewable memory, policy-as-code, $39/mo Pro.
• The gap nobody fills: humans and agents working in one shared, visual, navigable workspace.

"Everyone's building memory for agents. I'm building the place where humans and agents think together."

"The memory layer is already a commodity. The workspace is the moat."