Hi everyone. I’ve been using Continue for a few months now, mostly while building Airflow DAGs and dbt models. I have to say, I'm starting to feel like the whole category is... not getting worse, but just not getting *better* in a meaningful way anymore. The initial leap from having nothing to having Copilot/Continue was huge, but now the incremental updates feel like minor tweaks.
My main worry, which might be specific to my role, is that these assistants are still terrifying to use near production data pipelines. They’ll confidently suggest a change to a complex BigQuery SQL merge statement or a pandas transformation that looks right but has a subtle edge case. I can't trust it without a full review, which defeats the speed benefit. For example, it often misses idempotency in DAG tasks or suggests Python list comprehensions that could blow up memory on large datasets.
```python
# It might suggest this for a task...
def process_data(df):
return [expensive_function(row) for row in df.to_dict('records')]
# When I'd need something more like this for safety:
def process_data(df):
for chunk in np.array_split(df, 10):
yield expensive_function(chunk)
```
I find myself using it mostly for boilerplate—filling out standard docstrings, writing simple configuration YAML, or drafting common pytest patterns. The "big" features, like the upcoming Continue semantic search or automated refactors, sound cool, but I'm nervous they'll introduce bugs I can't easily see.
Is anyone else feeling this plateau? Are there specific patterns or use cases where you've found AI assistants still providing a real step-change in productivity, especially in data engineering? Or are we just in a period of consolidation where the tools are good but not revolutionary anymore?