With the arrival of Agentforce and the increasing reliance on Data Cloud, how should companies balance governance and the cleansing of legacy data to ensure that autonomous AI agents do not make incorrect decisions or expose sensitive data?

@Marcos Gomes

In my opinion, governance should come before autonomy. Agentforce is only as reliable as the data and permissions it is given.

A practical approach is:

• Clean legacy data first by removing duplicates, standardizing values, archiving obsolete records, and defining data quality rules.
• Implement strong data governance with clear ownership, validation rules, monitoring, and regular data quality reviews.
• Apply least-privilege access so AI agents only access the data they genuinely need. Sensitive fields should be protected with Field-Level Security, Permission Sets, Shield Platform Encryption, and Data Cloud data policies where applicable.
• Establish trust boundaries. Don't allow autonomous agents to perform high-risk actions without human approval. Critical decisions (pricing, financial updates, customer record changes, etc.) should include approval workflows or human-in-the-loop checkpoints.
• Continuously monitor and audit agent activity using logs, event monitoring, and feedback loops to identify incorrect decisions and improve prompts, policies, or data quality.

Ultimately, successful Agentforce implementations depend less on the AI model itself and more on the organization's data quality, governance framework, and security controls. Clean, governed, and well-permissioned data is what enables AI agents to make trustworthy decisions.