# Is GitHub Copilot worth paying for now that it's generally available?

Date: 2022-07-01 · Author: Robert Scales · https://guidetoartificialintelligence.com/en/learn/github-copilot-worth-paying-that-generally-available/

**Answer:** For many developers who write code most days, yes: at 10 US dollars a month, Copilot pays for itself if it saves you well under an hour of work each month. The honest way to decide is to use the free trial on your own projects and measure the time it saves, because the benefit varies a lot from person to person.

On June 21, GitHub made Copilot generally available to individual developers. "Generally available" means the technical preview is over: anyone can now install it, and it is a paid product. The price is 10 US dollars a month or 100 dollars a year, there is a free trial, and students with a verified status and people who maintain popular open source projects can use it at no cost. GitHub says it will start offering Copilot to companies later this year.

So the question many people are asking this week is a practical one. During the preview it was free and easy to leave running. Now it has a price, and a price deserves a calculation.

## What you are paying for

Copilot is an editor extension that works in Visual Studio Code, Visual Studio, Neovim and JetBrains IDEs. As you type code or a comment, it proposes what should come next: a single line, a whole function, boilerplate (the repetitive setup code most files need), or a unit test (a small program that checks that another piece of code works). You can cycle through several suggestions and accept, reject or edit each one.

Under the hood it runs on a model from OpenAI trained on a large amount of public source code. It does not run your code or check that a suggestion is correct. It predicts likely code from the context in your file, which is why its suggestions can be excellent, slightly off, or confidently wrong.

GitHub reports that more than 1.2 million developers tried the preview over twelve months, and that in files where Copilot is turned on, nearly 40% of the code in popular languages such as Python is now written by it. That figure describes how much code it produces, not how much time it saves, and those are not the same thing.

## A worked break-even calculation

The simplest test is a break-even point: how much time must Copilot save each month to cover its cost? You need only your own hourly value, whether that is your wage, your freelance rate or what your free time is worth to you.

| Your hourly value | Monthly cost | Time saved needed to break even |
|---|---|---|
| $20 | $10 | 30 minutes a month |
| $40 | $10 | 15 minutes a month |
| $80 | $10 | under 8 minutes a month |

The annual plan lowers the monthly cost to about $8.33, which shortens each row a little.

Now try it on a small, typical task. Say you need a function that reads a CSV file of customer orders and returns the total per customer. Time yourself writing it by hand once. Then, on a similar task, write only a descriptive comment and the function name, and let Copilot propose the body:

```prompt
# Read orders.csv (columns: customer, amount) and return a dict
# mapping each customer name to the sum of their order amounts.
def totals_by_customer(path):
```

Count the whole cycle: waiting for the suggestion, reading it line by line, fixing what is wrong and testing it. If the task took 12 minutes by hand and 5 with Copilot, you saved 7 minutes. Two or three tasks like that in a month already cover the cost at most hourly values.

The catch is the reading step. A suggestion you accept without understanding is not time saved; it is a bug you may pay for later.

## What the early research says

A GitHub research team studied this question in a paper presented this June. The researchers matched more than 2,000 survey answers from preview users with usage data from their editors. They looked at the acceptance rate, meaning the share of shown suggestions that a developer keeps, and found it was the best predictor of whether people felt more productive. The average acceptance rate in their data was 27%, and it varied widely between developers and over time.

Two points matter for your decision. First, the study measures how productive people feel, not a stopwatch result. Second, the wide spread means your experience may differ sharply from a colleague's. That is why the free trial is worth using seriously, on real work, before you pay.

> **Try it:** During the trial, keep a simple log for one week: for each task where you used Copilot, note the minutes you think it saved and any time spent fixing its mistakes. At the end of the week, multiply the net minutes by four and compare with the break-even table above.

## How do you get the most from it?

Give it context, because context is all it has. Copilot reads the file you are working in, so clear function names, a short comment that states inputs and outputs, and an example of the data format all improve its suggestions. A comment like "parse the date in DD/MM/YYYY format" works better than "parse date."

Use it where it is strongest: repetitive code, common patterns, tests, and syntax in a language you know less well. Be more careful with security-sensitive code, unusual business rules and anything involving money or personal data, where a plausible but wrong suggestion is costly. When a suggestion looks almost right, look at the alternatives before editing; one of them may be closer.

Finally, keep your own judgment in charge. Read every suggestion as if a new colleague wrote it, run your tests, and reject freely. Developers who treat Copilot as a fast first draft, rather than a finished answer, are the ones most likely to find that the 10 dollars a month is easily earned back.

## Sources

- GitHub Copilot is generally available to all developers: https://github.blog/2022-06-21-github-copilot-is-generally-available-to-all-developers/
- Productivity Assessment of Neural Code Completion: https://arxiv.org/abs/2205.06537
- Introducing GitHub Copilot: your AI pair programmer: https://github.blog/news-insights/product-news/introducing-github-copilot-ai-pair-programmer/