Systems of record are evolving rather than disappearing. The author argues that while there are claims that new agents or workflows are replacing traditional systems of record, the need for a reliable source of truth remains vital for enterprises. The crux of the issue is where the canonical truth resides, especially as automation increases. For example, different departments within a company might report varying figures for annual recurring revenue (ARR), complicating decision-making. If an automated agent is tasked with calculating ARR, it could struggle to determine which figure to trust if there’s no clear guidance on what data source is authoritative. Historically, systems of record like CRM and ERP provided clear homes for data across domains. However, the shift toward centralizing data in warehouses or lakehouses hasn't fully resolved inconsistencies. These data repositories often serve as retrospective mirrors rather than proactive tools for real-time workflows. The introduction of agents changes this dynamic by operating across multiple systems and taking action based on the data they gather. The author highlights Databricks as a player poised to be central in this new paradigm by developing these agents and facilitating their operations. Data warehouses and lakehouses are positioned to support these workflows, acting as a "truth registry" for the enterprise. They centralize data and maintain a semantic layer that defines metrics. However, the current systems are designed for human queries, not agent orchestration. Agents require explicit rules and conflict resolution mechanisms built into the data model. The author suggests that traditional systems like CRM and ERP will not vanish but will evolve into API-driven state machines that prioritize programmatic access. The focus will shift from human interaction to machine communication, emphasizing the need for clear contracts about data usage and its authoritative sources.