---
title: "Learn Python, or learn to drive Copilot? A beginner's honest fork in the road"
date: 2026-07-12T00:00:00Z
updated: 2026-07-09T11:54:40Z
tags: ["Copilot", "Python", "Career", "Getting Started", "Microsoft 365", "Excel"]
canonical: https://bianca.codes/blog/learn-python-or-learn-to-drive-copilot-a-beginners-honest-fork-in-the-road/
---

# Learn Python, or learn to drive Copilot? A beginner's honest fork in the road

_Two tabs open, four spare hours a week, and everyone insisting the other tab is a waste of your time - here's the no-hype version of the beginner's fork._

There are two tabs open. The left one is a Python tutorial, paused at the point where you typed `print("Hello, world")` and felt briefly powerful. The right one is a Copilot prompt box, cursor blinking, waiting for you to describe the spreadsheet problem that made you open both tabs in the first place.

You have maybe four spare hours a week for this. Both tabs are promising the same thing - that you'll stop doing the tedious parts of your job by hand - and everyone with a platform has told you, loudly, that the other tab is a waste of your time. The Python people say prompt-typers will never understand what they're running. The AI people say learning syntax in 2026 is like learning shorthand in 1995.

Both camps are selling something. Neither is describing your situation, which is: limited time, a real job, and a genuine question about where the hours pay off. So here's the no-hype version.

## The two tabs are offering different kinds of payoff

The honest difference isn't capability. It's the shape of the return curve.

Copilot pays off like rent: immediately, and again next week, for as long as you keep showing up. Python pays off like an asset: nothing for weeks, then compounding returns that change what kind of problems you can take on at all.

Most of the arguing online is people with one curve shouting at people with the other. The actual decision is about which curve fits the problems in front of you - so it's worth being precise about what each one buys, and where each one quietly stops.

## What driving Copilot buys you, and where it stops

The payoff starts this week. Describe the outcome you want - a formula that flags duplicate invoices, a summary of a 40-email thread, a first draft of the report nobody wants to write - and you get something usable in seconds. If you work in Excel, Word or Outlook all day, that is not a small thing. I've [already made the case](/blog/copilot-hot-take/) that opening Copilot beats reading takes about it.

And driving it well is a real skill, not a consolation prize. Writing a prompt that actually works means stating your inputs, your desired output, and your edge cases precisely. That's specification. It is, quietly, the hardest half of programming, and Copilot makes you practise it from day one.

Here's where it stops. **You cannot check what you cannot read.** The day Copilot writes you a formula that's wrong in a subtle way - and that day comes for everyone - you have two options: take it on faith, or rebuild the logic by hand until you find the disagreement. Without some grounding in how the logic works, your quality ceiling is "looks plausible", and plausible is exactly the failure mode. The limit isn't what Copilot can produce. It's what you can verify.

There's a quieter limit too: Copilot is strongest inside tools you already understand. It amplifies existing competence. It's much weaker at giving you competence you don't have.

## What learning Python buys you, and where it stops

The slow start is real, so let's not decorate it. Weeks of exercises that feel like homework, and a long gap between `print("Hello, world")` and anything your manager would recognise as work. Most beginners quit somewhere around list comprehensions, not because the concept is hard but because nothing they've built yet matters to them.

What you're actually buying, though, isn't syntax. It's a set of mental models: data as structures you can inspect, work as a sequence of transformations, error messages as information rather than punishment. Those models transfer everywhere - Power Query, DAX, even the way you structure a complicated workbook. They are also, not coincidentally, exactly what you need to judge whether generated code is right. If you want to see what the early payoff looks like in practice, [Part 1 of the Python for Excel Users series](/blog/python-for-excel-users-part-1-reading-and-filtering-your-first-real-dataset/) gets you from zero to filtering a real dataset in one sitting.

Where it stops: time, and maintenance. Every script you write is something you now own. Six months from now it will break on an edge case, and you'll be the only person in the building who can fix it. Python also punishes learning in a vacuum - without a real problem pulling you forward, it's just increasingly elaborate homework.

## The fork is drawn wrong

Here's the part both camps skip: the question assumes these are substitutes. They're complements, and they cover each other's weaknesses almost perfectly.

Python's classic beginner problem is the desert between the tutorial and the first useful script - the part where you know what a loop is but not why your dates won't parse. That desert is precisely what Copilot removes. Stuck beginners used to spend an evening on a missing bracket; now the assistant explains the error, in context, immediately.

Copilot's problem is the verification ceiling - and basic Python literacy is precisely what raises it. You don't need to write fluent pandas. You need to read a dozen lines of generated code and spot that it's summing the wrong column.

So the honest sequencing, if you're starting from zero with four hours a week: start with Copilot, this week, on a problem you actually have. Not exercises - the real report, the real inbox. Then, the first time it hands you something that matters enough to double-check, follow that thread into the basics. Learn to read before you learn to write. You'll learn faster than the tutorial route anyway, because every concept arrives attached to a problem you already care about.

Copilot didn't make learning Python pointless. It removed the worst part of learning Python and made the remaining part - judgment - more valuable. It is, accidentally, the best on-ramp to code the platform has ever shipped.

## Frequently Asked Questions

### **Do I need Python at all if I already know Excel well?**

Not necessarily. If your work fits in Excel and Power Query, deep skill there beats shallow Python. The signal to learn Python is hitting the same wall repeatedly: files too big, transformations too awkward, or the same manual process eating an hour a week. Until the wall shows up, sharpening the tools you use daily pays better.

### **Can I skip code entirely and just get better at prompting?**

You can go a surprisingly long way, especially inside M365 where Copilot works with tools you already know. But your ceiling becomes verification: for anything where a subtly wrong answer costs money or credibility, someone has to be able to read what was generated. If that someone isn't you, you're delegating judgment, not work.

### **Should I start with proper Python or Python in Excel?**

If your data already lives in spreadsheets, [Python in Excel](/blog/python-in-excel-cell-becomes-a-dataframe/) is the gentler entry - no environment setup, and your data is right there. Move to standalone Python when you need files, scheduling, or anything Excel can't host. The concepts transfer completely; only the plumbing changes.

### **How long before Python actually pays off at work?**

With a real problem pulling you forward and Copilot unblocking you, weeks, not months - the Python for Excel Users series gets you filtering and summarising real data in the first sitting. The old six-month estimate assumed you were learning alone from generic tutorials. That assumption is dead.

### **Won't AI make Python obsolete before I finish learning it?**

The syntax-recall part of the skill is already fading in value. The judgment part - knowing what to build, whether the output is right, and what the edge cases are - has gone up in value precisely because generation is now free. You're not learning Python to type it. You're learning it to supervise it.

## The way to think about it

Stop treating the two tabs as rival careers and treat them as one loop: Copilot gets you output today, and the need to trust that output is what makes the fundamentals worth learning. Pick problems, not tools - the real report, the real folder full of files - and let each problem tell you which tab it needs. If the work you're automating is file-shuffling rather than analysis, Monday's post on Power Automate triggers is that loop applied to a different tool. And in a couple of weeks I'm publishing a beginner's honest week inside Copilot - the discovery log version of this argument.

The fork in the road was never Python versus Copilot. It's whether you build judgment or rent it. One of those compounds.