Bob Burman, race car driver. View public domain image source here

The Best Developers Now Produce 46 Times More Than Average Ones

Cursor released the Spring developer report and I found this graph the most interesting.

It is showing that power users account for the largest share of AI lines, spend and token consumption.

Not a Medium member? Keep reading for free by clicking **here**.

That looks like a perfect Pareto distribution curve.

Pareto Type I probability density functions for various α -source Wikipedia

In other words it’s the 80/20 rule again, that is 80% of the AI code is written by 20%.

In fact going back to that graph, it might even be that just 10% of the users are writing 75% of the code.

This is quite phenomenal and backs up an argument that I made earlier in the year, that the future will look like Formula 1. Basically my argument is that the world we are entering into requires understanding how to operate at speed and that not everyone is going to be able to do it, just like not everyone is able to drive a Formula 1 car.

A Formula 1 driver processes information at a speed that would overwhelm most people. They’re reading telemetry, feeling tyre degradation, adjusting brake bias, and making overtaking decisions all within the same corner.

The p90 developers are doing something similar. They’re reading AI output at speed, making architectural calls in real time, and knowing when to throw everything away and start again. It’s a skill to be able to do this consistently.

The biggest evidence of this is in the next chart.

The 90th percentile developer is pulling further and further away from the median developer and the gap continues to grow overtime.

In fact if you shift up to the 99th percentile it is just staggering how much they are breaking away.

The 99th percentile developer is now producing 46 times more lines than the median active user and 15 times more pull requests.

This is quite incredible. This is also because the median developer in this will be writing far more code than if they weren’t using AI.

In terms of what this means from a raw number of lines adding per week, we can see that it has increased from 4,000 at the start of last year to over 8,000.

The online counter argument is that “well lines of code aren’t a measure of quality” and while there is merit to that, I think the more fascinating to me is those highly prolific developers. I certainly would hire someone at that 99th percentile right now, they have clearly learned.

If I look at my own Lines of code graph, you can see a similar upward trend since the start of 2026.

Last July I was doing days with roughly 2,200 lines of code edited.

This year I have a few days with over 50,000 lines of code edited and one day with over 70,000.

This reflects the fact that I am working on larger projects and more projects in parallel.

I certainly have been building more ambitious things this year but even with my largest weeks, I’m at 200,000 lines of code and the p99 are at 360,000.

I definitely see how that is feasible though, for me those high numbers come when you are deep in a project. But I often find that I need to then check in with the client and talk through what has been built and so that will lower down the average line count.

My counter argument to the lines of code being overrated

If I take my graph again, this last year has been extremely productive. It has resulted in dramatic performance across a number of domains, industries and clients for me.

Has everyone of those lines been perfect? No, absolutely not. Has the AI sometimes done stupid things? Yes, 100%. But is that graph above the best rough indicator I have as to the value that has been produced? Yes, it is and also the real world things like business revenue is the other one.

But across millions of developers, I fail to see how you can really measure anything apart from lines of code, as why would so many people be smashing out this many lines of code if they didn’t find utility in them?

It’s like how the number of questions on StackOverFlow could give you an indication of the total utility of the tool.

Apart from lines of code analysis there was also a few interesting charts around the pricing.

In the above graph the average request cost is $0.18 if you use Composer 2.5 and $1.57 if you use Opus 4.7.

Since Composer 2.5 has been released that is all I have been using and it has dramatically lowered the cost of development.

I like this next graph, it’s a bit buzzy to read but it’s the cost per intelligence chart.

Basically the goal in the above graph is to be at the top right. You want a cheap model that is really smart. Cheap is on the right and smart is at the top.

Composer 2.5 and GPT 5.5 are the winners of this by a long shot.

But here are some of the discussion points that I’d love to have in the comments with this post

The hiring question is the practical one. I said I’d hire a p99 developer right now, but the problem is how you’d actually find one. Do I just ask to see their Cursor Lines of Code chart? Maybe, but then you’d have to dive a bit deeper, and see how someone talks about their workflow. I’m actually hiring for a data scientist at the moment, so will try this out. If you are based in Auckland then feel free to add me on LinkedIn!

Another thing I’m wondering about. If the gap keeps growing at this rate, what happens to the median developer? They’re still productive. They’re writing more code than they were two years ago. The problem is that someone else is writing 46 times more. This creates a strange dynamic where everyone is getting better and most people are falling behind simultaneously.

The “will AI replace developers” debate may turn out to have been the wrong question. The right question might be something more random like whether excellent developers with AI will replace average developers with AI. At a certain point, if one person can do the work of ten, it makes it rational to go with the outlier right? Am I missing something here?

If you find the Power Law and Pareto distribution fascinating, then you will enjoy this book, which talks about Power Law’s in the context of Silicon Valley and Venture Capital.

View the book on Amazon here.

I’m an Amazon affiliate, I make a small commission if you make a purchase.

Before you go

Subscribe to my free Substack newsletter because you get the following:

• A brand-new article for executives on Sunday that’s only posted on Substack.
• Links to every Medium post I’ve written in the past week
• Book recommendations every week for you to spend your Audible credits on

What do you think? Is this turning into a Formula 1 world future, where we just have these superstar outlier developers? Or has that always been the case and this is just exaggerating things?