Most memory systems decide on the write path. An event happens, a model extracts what seems important, and the rest is discarded. That keeps the store clean. It also asks the write path to know, at ingest, what will matter in three months. A detail that looks irrelevant in a support chat may explain an escalation later. Once the raw event is gone, better retrieval cannot recover it.

The better pattern is the old data-lakehouse move: preserve what happened, then build the useful representations later. Summaries, embeddings, graphs, and memories become replaceable views over an immutable history. The same conversation can yield one set of facts today and another next year. Better models can re-index old history. New agents can use it in ways nobody planned at capture time.

The bullish case for incumbents is that AI makes systems of record more important, not less. Pipe the data into a general agent and skip the AI-native app. Claude becomes the front door. Salesforce still owns the CRM. Incumbents are also moving from storage into action: reviewing contracts, routing tickets, building campaigns.

The argument is partly correct. It is also incomplete. A contract is not the legal matter. A ticket is not the resolution. The job crosses applications, teams, and companies. A general agent can reach across tools, but access is not competence. Latency, inconsistent entities, and information held by other parties still block the full job.

Agents now read, write, and act directly on data through APIs, making the human-facing UI the weakest layer of enterprise software.

Traditional systems of record once derived defensibility from daily usage, muscle memory, undocumented workflows, and human-enforced data hygiene. Those advantages fade fast when agents become the primary users.