Two teams reporting different numbers for the same word
Active users mean one thing in the product dashboard, another in the board deck, and a third in the marketing report. Each definition was reasonable. None of the definitions were defined together. Every meeting that contains two of them wastes ten minutes.
Finding the divergence is not a search problem. The definitions live in query logic, dashboard filters, a spreadsheet formula, and somebody's head. None of them are labeled as definitions.
Deciding which should stand is genuinely contested. The product definition is often more accurate. The board definition is often more comparable to history. Picking either definition has consequences.
Divergence can also be legitimate. Two teams measuring different things that happen to share a name should rename rather than reconcile. Telling those cases apart is the real conflict.
How the agent is built in Agent Studio
Definition sources are what the agent needs access to, and they are scattered by nature. A PostgreSQL node covers query logic. A Notion node covers the metrics documentation. A Google Sheets node covers spreadsheet formulas. An HTTP Request node covers dashboard configuration.
The system prompt states what makes one definition preferable in your organization: accuracy, historical comparability, or the one the board already uses. It requires that the trade-off be named rather than assumed.
Prompt skills carry your canonical metric list, so the agent knows which names are shared.
A Start Scheduled flow assembles the definitions, and the Agent Chat node requests a comparison.
Notion receives a document per conflicting metric: each definition found, where it lives, what it produces, and a recommendation with the trade-off stated.
The agent never changes a definition. Reconciliation is a decision with reporting consequences. That decision belongs to whoever owns the metric.
A Slack node posts only genuine conflicts. Rename recommendations remain visible rather than being folded into the conflicts.
What this agent is built from
- System prompt: states what makes a definition preferable here, and requires the trade-off to be named.
- Prompt skills: carry the canonical metric list, so shared names are known.
- PostgreSQL (Data): supplies query logic. Notion, Google Sheets, and HTTP Request cover the rest.
- Agent Chat (Util): a scheduled flow assembles the definitions and requests a comparison.
- Notion (Integration): receives one document per conflicting metric, along with the recommendation.
- Boundary: the agent never changes a definition. The metric owner decides.
- When an ActionFlow is enough: If you only need to list where a metric name appears, build the workflow. Arguing which definition should win is the agent's job.
Frequently asked questions
Start building AI workflows
Create a free account, open a template or a blank canvas, and run your first ActionFlow.
Newsletter
Get product updates
New nodes, agents, and product notes. We send mail only when we have something worth opening.