What Our Copilot Data Actually Says About AI Coding Efficiency

There’s a lot of noise right now around AI-assisted coding. Some people swear it’s a massive productivity boost. Others think it just adds another layer of things to review. I personally, as a non-coder, find it very useful. It helps me write some quick JS, or a PowerShell script if I want to test or automate something quick and dirty. But what about developers? People whose job is to write clean, efficient, robust code that shapes the product that runs the business? Luckily, I didn’t have to go too far to find the answer. I work for SquaredUp, that sells a dashboarding software that runs our business. Obviously, we have many talented developers who write this software that our customers praise and love. What better than to run some numbers internally on how they like to use GitHub Copilot?

So that’s exactly what I did. I built a SquaredUp dashboard to plot these metrics.

What did I find? Let’s take a look!

The headline numbers

• 4,057 times in the last 28 days they accepted the Copilot suggested code resulting in 43,312 lines of added code
• 27.39% overall acceptance rate

At first glance, this looks like solid usage. People are clearly using the tool, and it’s generating a lot of code.

But one number stands out immediately:

Only about a quarter of Copilot’s suggestions are accepted.

That’s where things get interesting.

Code adoption has a range

Here’s a breakdown of how our devs have been using GitHub Copilot recently:

There are days where thousands of lines of code are suggested. But the acceptance rate tells a different story. In reality, developers are:

• Accepting some suggestions as-is
• Tweaking many (most) of them
• Ignoring a large portion entirely

So while Copilot is great at getting something on the screen quickly, that “something” isn’t always what you actually need.

I reached out to Josh - who had the least acceptance rate of all 12 (about 10%) - said this:

“It just never does things how I would have done them! 🤣” he also went on to add, “Possibly it's low because I usually use Copilot to do research and discuss approaches.“

Clearly Josh prefers his way of writing code, tried and tested by his years of coding experience. But he mentions that he values the AI research in context of the code and uses AI as a “sparring partner” for ideas and alternatives.

I also spoke with Alan, on the other edge of spectrum - who had the highest acceptance rate for suggested code (40%). He had this to say:

“It's probs that high because I use the autocomplete ability a lot where it suggests code changes, rather than asking it questions.” but also said, “Saved me a bunch of research time, but I had to explain a lot of the details of our framework and build so maybe it balanced out a bit 😅”

Again, highlighting the convenience of using AI research from within the IDE in context of your code. But it also suggests that the efforts it takes (prompt engineering!) to get it to write the right code still remains a struggle.

Despite the mixed reactions and personal preferences - the fact does remain that at the end of the day, close to half a million of lines of code was added in a month and honestly, that’s impressive. It just isn’t the same as “writing code for you.” (yet)

Here’s how you can create your own dashboard

It is easy to create your own dashboard like this. The dashboard above queries the REST API endpoints for Copilot usage metrics - GitHub Enterprise Cloud Docs API for organization level usage metrics and the REST API endpoints for Copilot user management - GitHub Docs API for license information.
There are 2 ways to query an API endpoint in SquaredUp: the GUI-driven WebAPI data source and the more flexible PowerShell data source. Normally, I’d use the WebAPI tile, but I’m querying the 28-day report API and it doesn’t return the metrics in the response payload - it provides download URLs that has this data in it. So a further call to those URLs is needed to get the data. This is easily possible with a PowerShell script so that’s what I opted for.

Here’s a sample script (made with GitHub Copilot 😉):

# Hardcoded values for quick test, hide these
$token = "<token>"
$org = "<org>"
$apiVersion = "2026-03-10"
$baseUrl = "https://api.github.com"
$url = "$baseUrl/orgs/$org/copilot/metrics/reports/organization-28-day/latest"
$headers = @{
Accept = "application/vnd.github+json"
Authorization = "Bearer $token"
"X-GitHub-Api-Version" = $apiVersion
}
$response = Invoke-RestMethod -Method Get -Uri $url -Headers $headers
$reports = @()
foreach ($downloadUrl in $response.download_links) {
$reports += Invoke-RestMethod -Method Get -Uri $downloadUrl
}
$reports.day_totals

Happy dashboarding!