Five verified 2026 differences between Excel Copilot Python (cloud, Anaconda, the sheet) and a portable .py file. For bank, insurance, finance, and ops analysts who live in Excel by day.
Excel still owns the day. In August 2026 Microsoft added a Copilot Python skill for Insiders: type @python-in-excel and Copilot can insert editable, refreshable Python formulas in the workbook. Copilot's everyday data Q&A already uses Python behind the chat. That is real Python — and it is not a .py file you own.
This listicle is for bank, insurance, finance, and ops analysts who live in Excel by day and want to understand Copilot's code at night. Hong Kong is rolling out 全民AI literacy and prompt classes at scale — useful, and a different product. Literacy teaches you to ask. A short hands-on course teaches you to run Python outside Excel so the next export still works when Copilot is offline or IT has not rolled the Insider skill to your build.
If you want a one-evening how-to — export one sheet, plot one PNG — that is a sibling piece: Make one chart tonight: from Excel to a Python file. This article is the five differences, so you know when the grid is enough and when you need a real file.
Python in Excel puts Python in a worksheet cell. You type =PY (or Formulas → Insert Python). Microsoft's introduction is explicit: calculations run in the Microsoft cloud, with a standard Python and a core Anaconda library set. You need internet. You do not need a local interpreter. A home venv and those pip packages do not apply inside those cells.
The August 2026 Excel blog lists "Copilot in Excel: Python skill (Insiders)": analyse and visualise with Python in Excel, create editable, refreshable Python formulas, and start using @python-in-excel. A separate Copilot path — data Q&A — uses Python-based analysis on tables and table-like ranges without asking you to write code. You can expand a commented code block to see what ran. Advanced analysis mode opens a new sheet and can insert the underlying Python as a refreshable cell.
Portable Python runs on the machine in front of you. No Copilot pane. No Azure container. Offline CSVs still open. Set up Python after work in 30 minutes is the evening that makes that true.
Grid is enough when the deliverable is a shared workbook and colleagues will never leave Excel. A .py file is required when you must run the same logic on a laptop without Microsoft 365, on a server, or after the VPN drops.
Python in Excel ships a core Anaconda distribution. By default you get pandas, NumPy, Matplotlib, seaborn, and statsmodels. You can import further libraries from Anaconda's published list (scikit-learn is on that list). Microsoft's data-security article is the real constraint: the container has a curated set of secured libraries; the Python code has no network access and no access to files on your computer. That is not pip install whatever. It is not uv add. You cannot reach PyPI from inside the cell.
On a real interpreter you choose the installer (pip or uv), pin versions, and keep a requirements.txt next to the script. That is how your first program that uses a library actually works — the package is on your machine.
Grid is enough for pandas-on-a-table and a seaborn chart the workbook can refresh. A .py file is required when you need a package Anaconda did not ship, a private wheel from work, or a pin you can reproduce in 2028.
read_csvMicrosoft's getting-started article is blunt. All data you process with Python in Excel must come from the worksheet or through Power Query. To protect security, pandas.read_csv and pandas.read_excel are not compatible. You talk to Excel with xl("A1"), ranges, tables, and Power Query names. Copilot Q&A only works on tables or table-like ranges; unstructured data is not supported. The Python code cannot open a network URL or a file on disk.
Portable Python starts from files. pd.read_csv("export.csv") is the normal first line after File → Save As → CSV. APIs, a folder of month-end dumps, and last quarter's extract all work because the process can see the disk.
Grid is enough when the numbers already live in the workbook you have open. A .py file is required when the source is a CSV on the desktop, a directory of files, or anything Power Query has not already pulled in.
Copilot's default Q&A does not modify the workbook. You can insert a static table or image; Microsoft says those inserts are not refreshable. Advanced analysis mode is the path that inserts a refreshable Python cell you can edit.
Even then, the code lives in =PY cells, in row-major calculation order, inside a .xlsx that typically sits on OneDrive or SharePoint. Excel has version history. That is not git log on a script. There is no .py for a colleague who does not open Excel, and no scheduler that runs the file at 07:00 without Excel. Copilot chat history (also in the August 2026 update) helps you revisit prompts. It does not replace a repository.
A portable script is a file. You commit it. You diff it. You re-run last month's CSV. The artefact survives a new laptop and a new tenant.
Grid is enough for a living workbook your team refreshes in Excel. A .py file is required when audit, Git, or a machine without Excel is part of the job.
Python in Excel needs a paid Microsoft 365 licence that includes the desktop apps. Microsoft's availability page: free consumer and perpetual licences do not support it; device-based and shared-computer-activation licences do not either. You need internet; it is a connected experience. An optional Python in Excel add-on buys premium compute and extra calculation modes (manual / partial / automatic). Copilot-with-Python is a further gate: the licensing FAQ says you can use it if you meet Copilot availability requirements. The August 2026 Python skill is labelled Insiders — your bank or insurer build may not have it yet.
CPython on your laptop does not bill Microsoft 365. You download an interpreter, install packages, and run files. Copilot is optional.
Grid is enough when your employer already pays for Microsoft 365 and Copilot, and the workbook is the product. A .py file is required when you change jobs, lose Copilot, or need the script to outlive the tenant.
.pyStay in Excel when:
.xlsx, not a terminal.=PY cells are the hand-off.Leave Excel when:
read_csv, local files, or APIs.Daytime Excel is not a failure. Night-time portable Python is how you stay competitive in the seat you already have. 全民AI will keep teaching prompts. This piece is why a short hands-on course exists: so you can read Copilot's Python at 22:00 and still run a file that is yours.
The enrollable path is Introduction to Python Programming — 19 published short lessons for after work. Check that course page for the current enrollment window; this article will not freeze a date. For pacing, see Learn Python after work: 19 short lessons. For the Hong Kong classroom contrast, see large AI classes vs short hands-on lessons. Then come back to the one-chart evening with a .py that is actually yours.
@python-in-excel, editable refreshable Python formulas (25 August 2026).@python-in-excel trigger are taken from Microsoft's blog, not this recap.xl(), data from the sheet or Power Query, pandas.read_csv / read_excel not compatible, row-major calculation order.xl() only.