Atomic Role Decomposition for AI-Powered Operations
A framework for breaking work down into its smallest useful units — tasks, workflows, and decisions — and reassembling them so humans and AI each handle what they do best.
The problem with automation-first thinking
Most automation projects start by asking what a tool can do. The better question is what the work actually requires. Until you decompose a role into its atomic parts, you cannot know which parts should stay human, which should become deterministic automation, and which can be handled by an AI agent.
The decomposition stack
Every role can be broken into three layers:
- Tasks: The smallest units of work — answer an email, approve a request, reconcile a report.
- Workflows: Sequences of tasks that produce an outcome — onboard an employee, process a payroll exception, resolve a compliance flag.
- Decisions: Points where ambiguity requires judgment — investigate a discrepancy, approve a non-standard arrangement, escalate an ER issue.
Once the work is mapped at this level, you can route each unit to the right owner.
Ownership model
I use a four-category ownership model to decide who or what does the work:
- Automation: Deterministic, rule-based tasks with no ambiguity.
- AI agent: Reversible, language-based, or ambiguous tasks where an LLM adds value.
- Human: Judgment, relationships, exceptions, and accountability.
- Hybrid: Human confirms or refines output generated by a system or agent.
The rule is simple: try deterministic automation first. Only use AI where the task is too ambiguous for hard rules but still reversible. Never automate accountability.
Why this matters now
AI is making it possible to restructure work at a level we have not seen before. But most organizations will waste the opportunity because they automate before they redesign. Atomic role decomposition is the prerequisite: it forces you to understand the work before you change who or what does it.