Skip to main content

Copilot Best Practices and Limitations

Preview

This feature is currently available in a private preview and may not be accessible to all customers.

Coalesce Copilot is an actively developed feature designed to accelerate your data pipeline development. While we're continuously improving the experience, there are important considerations to keep in mind.

Copilot Is Evolving​

As with any AI-powered tool, Copilot is under active development. We regularly ship improvements based on user feedback and testing, but you may occasionally encounter:

  • Unexpected node configurations or SQL outputs.
  • Intermittent service interruptions.
  • Changes to available capabilities as we refine features.

Best Practices​

Always Review and Verify​

  • Examine all generated SQL, node configurations, and transformations before deploying to production.
  • Test changes in a development environment first.
  • Validate business logic matches your requirements.
  • Check that generated code follows your organization's standards.
  • Review column mappings and transformations for accuracy.
  • Verify join conditions and relationships.
  • Ensure business logic aligns with requirements.
  • Test generated SQL syntax in your target platform.
  • Confirm Node Types match intended use, such as Dimension versus Fact.
  • Check that business keys are properly set.
  • Verify system columns are preserved.

Use Version Control Checkpoints for Safety​

  • Create a checkpoint before starting Copilot sessions.
  • Review all changes in version control before deployment.
  • Roll back easily if Copilot makes unwanted changes.

If You Encounter Issues​

  • If Copilot stops responding or produces an error, wait 1-2 minutes before trying again.
  • Start a new conversation if the issue persists—sometimes a fresh context helps.
  • Complex requests may take longer to process; simpler, focused prompts often work better.
  • Report persistent issues to our support team so we can investigate.

Provide Clear Instructions and Iterate​

  • If results aren't what you expected, rephrase your prompt.
  • Add more specificity or context.
  • Break requests into smaller steps.

What Copilot Does Well​

  • Generating boilerplate transformations and staging layers.
  • Converting SQL queries into node chains.
  • Adding columns and updating node configurations.
  • Explaining existing node logic and dependencies.

Where You Should Be Cautious​

  • Complex multi-node dependencies with intricate business logic.
  • Migrations from legacy systems (always validate thoroughly).
  • Production-critical transformations without testing.
  • Interpreting domain-specific business rules.

Team Collaboration​

  • Share Copilot-generated Nodes for peer review.
  • Document business context that Copilot can't know.
  • Use Copilot for scaffolding and the Coalesce UI for fine-tuning.
  • Track time saved on repetitive tasks.
  • Monitor accuracy of generated transformations.
  • Identify patterns where Copilot excels or needs guidance.

Security and Data Governance​

Understand Copilot's Scope​

  • Copilot sees your Workspace metadata such as Node names, columns, and schemas.
  • Copilot doesn't have access to actual data in your tables.
  • Copilot respects your Workspace permissions and Storage Locations.

Avoid Sharing Sensitive Information in Prompts​

  • Don't paste proprietary business logic that should remain private.
  • Use generic examples when asking questions about patterns.

Learn with Copilot​

Ask questions to deepen your understanding:

  • "How should I structure slowly changing dimensions?"
  • "What's the difference between Persistent Stage and View for this use case?"
  • "Can you explain the join logic you created?"

Copilot Limitations​

Workspace-Level Operations​

Copilot can't:

  • Create source Nodes.
  • Create or delete entire Subgraphs. It can only add or remove Nodes from existing Subgraphs.
  • Modify Workspace settings or configurations.
  • Manage Environments or deployments.

External Data Operations​

Copilot can't:

  • Query your data warehouse to preview actual data.
  • Create new Storage Locations or database connections.
  • Run actual data transformations. It only builds the DAG.

Advanced Customization​

Copilot can't:

  • Set advanced Node configuration options.
  • Create Custom Node Types.
  • Modify Package definitions or install new Packages.
  • Change Workspace-level templates or Macros.

Data and Metadata Awareness​

Copilot:

  • Knows your table schemas and metadata.
  • Doesn't know the actual data in your tables.
  • Can't automatically detect data quality issues without your description.
  • Can't recommend optimizations based on query performance.