When to use what
Go by the question
Section titled “Go by the question”| The task | Use | Why |
|---|---|---|
| “What was net revenue by region last quarter?” | SEMANTIX | Uses certified metrics and the definitions your team agreed on, read-only |
| “What does this metric actually measure?” | SEMANTIX | The metric’s definition, run history, and certification state are the answer |
| “Why did revenue drop in EMEA?” | SEMANTIX | Anomaly breakdowns and drift triage, with lineage and evidence |
| “Which table has customer data, and what is in it?” | SEMANTIX | Registry search, entity descriptions, lineage, and trust level |
| “Ad-hoc query against my data, with guardrails” | SEMANTIX | Read-only enforcement, 200-row cap, and the result is reproducible |
| “This number looks wrong — can I check it?” | SEMANTIX | Every answer keeps its question, SQL, and result in the thread |
| “Add this metric to the catalogue” | SEMANTIX | File a metric request; an admin approves it into the Registry |
| “Refactor this module and fix the failing test” | Coding agent | Code edits, running tests, repo-wide changes |
| “Add an endpoint to the backend” | Coding agent | Writing and editing code in a real repository |
| “Wire this new service into the platform” | Coding agent | Integration work with file and command access |
| “Research a topic with no connection to our data” | General chat | No governed context needed; do not manufacture one |
| “Call our internal API and summarise the result” | SEMANTIX open mode | A custom agent with the MCP tool attached, explicitly outside scope governance |
The rule of thumb
Section titled “The rule of thumb”Questions about your data go to SEMANTIX. Changes to your code go to a coding agent.
If a task is about understanding something you already have — data, metrics, definitions, trends — it belongs in the governed context. If it is about changing something you own — code, infrastructure, configuration — a coding agent is the right tool.
Three situations worth pausing on
Section titled “Three situations worth pausing on”You have data but no governance yet. Start with the quickstart: one provider, one scope, one connector, then introspect. Do not expect a large governed catalogue on day one — the value comes from the metrics you certify, not from the tables you register.
The question is one-off and nobody will ask again. A general assistant is faster. Ask SEMANTIX only when the answer needs to be defensible or will be reused — otherwise you are paying setup cost for a single question.
The answer needs to outlive the conversation. This is the strongest argument for SEMANTIX. A chat answer is a moment; a certified metric is an asset. If the question will be asked again next month, it belongs in the Registry, not in a transcript.