Mapping existing human processes is useful for context, but forcing AI agents to follow them is a mistake. Instead, define clear goals and constraints, then let the agents determine the most efficient path, which will likely be non-human and more effective.
While a single source of truth is the ideal, large organizations should aim for an interconnected "mesh" or "lattice" of different data sources. AI agents can then traverse this mesh, identify discrepancies, and even request human help for reconciliation, which is a more realistic model for complex enterprises.
The rapid pace of AI advancement requires designing systems with the assumption of frequent, fundamental change. This means avoiding attachment to current workflows, even recently successful ones, and being culturally ready to reimagine everything from first principles on a regular basis.
Instead of viewing governance functions like legal and HR as barriers, truly AI-native companies treat them as transformation partners. They collaborate to design enabling policies and guardrails that unlock the ability to deploy powerful AI agents safely and at scale, making it a competitive advantage.
Granting full autonomy to AI agents from day one is reckless. A safer, more effective approach is a laddered model: start with agents in an "Observation" role, then let them make "Suggestions," then "Act with Approval," and only then grant full autonomy within specific, defined boundaries.
As AI automates tasks, the critical human role shifts from execution to ownership. This creates a new management discipline where every employee, not just traditional managers, must be accountable for the measurable goals and results of the AI systems they deploy, asking "who owns this?" before "can AI do this?".