Copilot Agent does not work reliably with Jupyter notebooks in VS Code. It can sometimes edit cells, but the behavior is inconsistent. The Agent often produces changes that don’t match what it says it will do, edits the wrong cell, removes parts of code or breaks the notebook structure. Cant handle at all long cells,( looks like the tool does not support targeted edits, only whole cell replacement).This becomes even more unreliable in notebooks that contain both Python and SQL cells.
The main issue seems to be that the notebook editor does not expose a proper model for the Agent to work with. Because of that, the Agent appears to guess cell boundaries and apply text edits to something that isn’t a normal text document. This leads to unpredictable results and makes it hard to use the Agent for any real notebook‑based workflow.
In many data engineering and analytics environments, notebooks contain Python, SQL, markdown, and other metadata. These cannot be converted to .py files because .py cannot represent SQL kernel cells or other notebook‑specific structure. Tools like Jupytext also cannot round‑trip mixed‑language notebooks, so using a script file as a workaround is not an option.
For these reasons, .ipynb has to remain the source of truth. To make Copilot Agent usable in this context, it needs proper support for notebooks: the ability to read cells, modify them reliably, insert or delete cells, and work with notebooks that contain more than one language.
I’m requesting first‑class, reliable Copilot Agent support for .ipynb notebooks in VS Code so the Agent can be used in real data workflows. I would expect that to work from the first publication, but it looks like it needs to be addressed while in production
Copilot Agent does not work reliably with Jupyter notebooks in VS Code. It can sometimes edit cells, but the behavior is inconsistent. The Agent often produces changes that don’t match what it says it will do, edits the wrong cell, removes parts of code or breaks the notebook structure. Cant handle at all long cells,( looks like the tool does not support targeted edits, only whole cell replacement).This becomes even more unreliable in notebooks that contain both Python and SQL cells.
The main issue seems to be that the notebook editor does not expose a proper model for the Agent to work with. Because of that, the Agent appears to guess cell boundaries and apply text edits to something that isn’t a normal text document. This leads to unpredictable results and makes it hard to use the Agent for any real notebook‑based workflow.
In many data engineering and analytics environments, notebooks contain Python, SQL, markdown, and other metadata. These cannot be converted to .py files because .py cannot represent SQL kernel cells or other notebook‑specific structure. Tools like Jupytext also cannot round‑trip mixed‑language notebooks, so using a script file as a workaround is not an option.
For these reasons, .ipynb has to remain the source of truth. To make Copilot Agent usable in this context, it needs proper support for notebooks: the ability to read cells, modify them reliably, insert or delete cells, and work with notebooks that contain more than one language.
I’m requesting first‑class, reliable Copilot Agent support for .ipynb notebooks in VS Code so the Agent can be used in real data workflows. I would expect that to work from the first publication, but it looks like it needs to be addressed while in production