Grade Cursor on a real data science project
A data scientist tested Cursor on a real project: weak on pandas, data cleansing and scraping, strong at prototyping. Task: Mike Woodward evaluated Cursor (using Claude) on a real Python data science project involving English football data. He prompted for code generation, then manually reviewed and fixed outputs, particularly for pandas dataframe manipulations where generated code failed. The workflow involved grading specific tasks like prototyping, scraping, and data cleansing.; tools: Cursor. Human role: Mike Woodward manually reviewed and fixed the generated code, including a pandas query for clubs appearing only as home teams.; limitations: The free tier exhausts tokens quickly, necessitating a paid subscription for meaningful evaluation. Cursor is not a notebook environment, which hinders data science workflows. It deleted data files during GitHub integration and struggles with technical code accuracy, requiring strict human verification..
Tools
Published on Agent Work Atlas:
Overview
A data scientist tested Cursor on a real project: weak on pandas, data cleansing and scraping, strong at prototyping.
Workflow
Mike Woodward evaluated Cursor (using Claude) on a real Python data science project involving English football data. He prompted for code generation, then manually reviewed and fixed outputs, particularly for pandas dataframe manipulations where generated code failed. The workflow involved grading specific tasks like prototyping, scraping, and data cleansing.
Reported result
Cursor scored A for prototyping, documentation, and code completion, but C for pandas manipulation and data cleansing, and D for web scraping and getting started. It generated Streamlit apps rapidly but produced incorrect or non-functional code for complex dataframe queries, requiring manual rewriting.
Human role
Mike Woodward manually reviewed and fixed the generated code, including a pandas query for clubs appearing only as home teams.
Limitations
The free tier exhausts tokens quickly, necessitating a paid subscription for meaningful evaluation. Cursor is not a notebook environment, which hinders data science workflows. It deleted data files during GitHub integration and struggles with technical code accuracy, requiring strict human verification.
Try it
Act as a data scientist evaluating Cursor for a Python project. 1. Set up a small dataset (e.g., CSV) and define a specific pandas manipulation task (e.g., filtering rows based on complex conditions). 2. Prompt Cursor to generate the code. 3. Run the code and verify results against manual expectations. 4. If it fails, note the error and attempt a fix via prompt; if it still fails, write the code manually. 5. Repeat for a Streamlit UI prototype task. 6. Record time taken and accuracy for each step. Do not trust generated code without execution and verification. Output a scorecard comparing AI-generated vs. manual effort.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.