What Happens in Elaboration
Elaboration is where structured work begins. Before any execution, teams define the intent — what needs to be accomplished, why it matters, and what success looks like.
This is not a requirements document. It is a living artifact that the team refines collaboratively, with the AI agent acting as a thinking partner rather than a passive recorder.
The output of elaboration is a clear, decomposed plan with verifiable completion criteria for every piece of work.
Key Activities
- Define the intent: Articulate what the initiative aims to achieve in plain language. No jargon, no implementation details — just the goal.
- Establish completion criteria: For every piece of work, define what "done" means in terms that can be verified. Not "looks good" — measurable, testable outcomes.
- Decompose into units: Break the intent into discrete, independently deliverable units of work. Each unit should be small enough to complete in a single focused session.
- Identify dependencies: Map which units depend on others. Build a directed graph that determines execution order.
- Select collaboration mode: Decide whether each unit needs supervised, observed, or autonomous execution based on risk and complexity.
How the AI Agent Behaves
During elaboration, the AI agent operates in a collaborative planning role. It asks clarifying questions, proposes decompositions, and challenges vague criteria.
The agent does not proceed to execution until completion criteria are specific and verifiable. This is the first quality gate — if the plan is vague, no amount of execution quality can save it.
Across Domains
| Domain | Elaboration Looks Like | |---|---| | Software | Feature intent, user stories, acceptance criteria, unit breakdown with dependency graph | | Marketing | Campaign brief, target audience definition, channel strategy, deliverable list with success metrics | | Operations | Migration plan, risk assessment, rollback criteria, phased implementation checklist | | Research | Research question, methodology selection, data source identification, analysis framework |
Outputs
- Intent document: The "what and why" — a concise statement of purpose with context
- Unit specifications: Discrete work items with titles, descriptions, and completion criteria
- Dependency graph: Which units must complete before others can begin
- Mode assignments: Supervised, observed, or autonomous for each unit