# Agentic actions In **Agentic** mode the assistant does not just propose SQL, it can run queries, read the results, and take the next step on its own: building a query, fixing one that errors, drafting a report, or suggesting schema improvements. Switch modes from the toggle in the chat composer. ## Build a query, then a report Ask for an analysis and the assistant runs the queries it needs, reads the rows, and assembles a written answer grounded in the real data.  ## Fix a failing query Paste or run a query that errors and ask the assistant to fix it. It inspects the schema and the error, then proposes a corrected query.  ## Suggest schema improvements Ask the assistant to review a table and it can suggest indexes and other improvements based on the table's structure and how it is queried.  ## Safety Agentic actions respect your connection settings. On a [production](https://docs.snoutdata.com/connections/security) connection the destructive-statement guardrail still applies, and writes are only ever made when you allow them.