I’ve been using Tabnine Pro for several months now, primarily for Python and JavaScript development, and overall, I’ve been impressed with its flow integration and code completion. However, I’ve noticed a consistent and significant drop in quality when working with SQL, whether in `.sql` files or within string literals in my application code.
The suggestions often feel generic or even syntactically incorrect. For example:
- It struggles with context-aware column name completion within `WHERE` or `JOIN` clauses.
- It frequently suggests outdated or incorrect table aliases.
- Complex window functions or CTEs are rarely completed correctly, often resulting in broken syntax.
This is particularly noticeable compared to its performance on general-purpose languages. It raises a question about the training data and fine-tuning. Given SQL's structured nature, one would expect strong performance.
I'm curious if others in the community have had a similar experience. Specifically:
* Is this a known limitation of the underlying model, or a configuration issue?
* How does it compare, in your experience, to other AI-powered tools that offer SQL-specific features?
* Have you found any effective workarounds—perhaps specific prompting patterns or settings—to improve its SQL suggestion accuracy?
I believe discussing this openly can help set realistic expectations for new users and potentially guide feature prioritization for the Tabnine team.
—lucas