1032: Agents Need 10x More Data Than Humans, with Salesforce’s CDO Michael Andrew

Episode Summary

Executive Summary: Salesforce CDO Michael Andrew argues that the AI era turns data teams into builders for both humans and agents. Because agents lack business context, organizations must create trusted, well-described data and unstructured context at much greater scale, while using zero-copy architectures and unified profiles to activate data without expensive migrations.

Main Topics: Agents as new data consumers (Priority: 5/5): Andrew explains that data teams now serve two customer types: human employees and AI agents, with agents requiring explicit business context because they do not know the organization’s terminology or processes. Data quality and trusted context (Priority: 5/5): He emphasizes that agents need highly reliable, well-labeled, and harmonized data because they often act without uncertainty; bad data directly degrades agent performance, cost, and outcomes. Salesforce’s Data360 transformation (Priority: 5/5): The discussion covers how Salesforce used Data360 to unify fragmented customer data across multiple systems and create a real-time harmonized view for sales, service, marketing, and products. Zero-copy data activation (Priority: 4/5): Andrew explains that modern architecture should avoid unnecessary data movement. Instead, companies should read data where it lives and activate it across workflows, warehouses, and applications. Agentic enterprise operating model (Priority: 4/5): He frames the future as an agentic enterprise in which every function uses agents, requiring coordinated planning between AI use cases and the data needed to support them. Skills for data practitioners (Priority: 4/5): Andrew advises data professionals to use multiple models, test prompts with varying context, and learn evaluation design so they can build intuition about how agents behave. Meditation and managing artificial minds (Priority: 2/5): In the closing segment, he recommends meditation as a personal discipline to cultivate calm and awareness while working with AI systems.

Key Arguments: Agents are now customers of the data team, not just internal tools, because they need structured context to operate effectively. Agents require far more trusted data than humans because they do not understand the business unless the business teaches them through data. A small number of skilled humans plus agents can outperform larger teams by multiplying data engineering and analysis capacity. The real value is not raw data storage but activating data in workflows where business users can use it. Zero-copy architectures are preferable because moving data is expensive and often unnecessary for most use cases. Data quality ROI is now measurable because one can compare how an agent performs with clean versus poor context. Data and AI must move in lockstep; AI strategy should start from the use case and work backward to the data required. Unstructured data such as documents, notes, and slides is becoming as important as traditional metrics and forecasts for agentic workflows. Practitioners should learn by experimenting across models, prompts, and evaluation methods rather than assuming deterministic outputs.

Data Points: Fragmented customer data streams: 650+ - Salesforce’s customer data was spread across more than 650 data streams before unification. Fragmented customer profiles: 266 million - Number of fragmented customer profiles Salesforce had to reconcile across systems. Unique individuals after resolution: 141 million - Profiles were collapsed into a single unified view of unique individuals. Timeframe of cleanup: 4 years - Period over which Salesforce resolved and unified the fragmented customer profiles. Salesforce internal scale: About $50 billion per year - Andrew describes Salesforce as a company operating at massive global scale, underscoring the importance of accurate data. Employees: 85,000 - Approximate number of Salesforce employees who depend on reliable data systems. Usage increase with agents: 6x - Customers documenting agents early saw six times the usage of Salesforce compared with human-only workflows. Potential productivity multiplier: 5-10x - Andrew says one great data engineer with five agents can produce the output of a team of five or 10. Zero-copy network usage: 90% of use cases - He says most use cases should read data in place rather than move it. Historical data system count: Multiple CRM instances plus Snowflake, Google, and Amazon - Describes Salesforce’s fragmented stack before Data360 integration. Data flow scale: Hundreds of millions of emails and messages yearly - Data360 is used to drive large-scale marketing communications. Personal meditation practice: 22 years - Andrew says he has maintained a deep contemplative practice for 22 years.

Pivotal Quotes: "We now have customers that are the agents." — Michael Andrew: Explaining how the role of data teams has changed in the agentic era. "I'm almost at like, I need to produce 10 times volume of trusted data than I did before agents." — Michael Andrew: Describing the rising data demand created by AI agents. "We’re realizing you can now quantify the ROI of the quality of the data in a way you never could before." — Michael Andrew: On how agent performance makes data quality value measurable.

Implications: Organizations must treat data as operational fuel for agents, not just analytics input. The winners will unify trusted context, activate data without costly movement, and build evaluation-driven AI workflows that can scale across every function.