In the Beginning was the Command Line is a famous essay by Neal Stephenson, in which he argues against the Disneyfication of computing represented by Graphical User Interfaces (GUIs). And I do have fond memories of programming using nothing but a text editor (VIM) and the command line. Over the years though Integrated Development Environments (IDEs) became more and more popular. And today most programmers are using something like Visual Studio or more recently Cursor.

But IDEs are still code centric and code is going away. Well not away entirely but it will become increasingly hidden just as machine code has been hidden from us for a long time. I grew up at just the right moment in time where machine code and assembly language was still something you learned and occasionally used directly because computers were slow and if you wanted to write a game, core graphics loops had to be hand written in assembly. Since then entire generations of programmers haven’t seen assembly or machine code. It hasn’t gone away entirely but it exists at a layer that is just there and works. Now the same will happen with higher level code. 

Coding agents are doing to source code what compilers did to machine code: push the code below the interaction surface.

This creates an interesting challenge for what the AI native IDE should look like. All one has to do is look at recent iterations of Cursor to see how rapidly this is changing. Cursor started as a fork of Visual Studio (Microsoft open source!) and was immediately recognizable as such. The most recent version on the other hand has an entire section for just dealing with coding agents that looks nothing like a traditional IDE.

The question then is what should the right environment look like? I believe these will be Intent-based Collaboration Environments. In the “nothing is truly new” department it is worth remembering that the idea of intent based programming goes back to at least the 1990s and Charles Simonyi and others even started a company called “Intentional Software” in 2002. They had the right idea but were simply a bit too early. Now we have coding agents that can turn intents into code.

So what would I want to see in an Intent-based Collaboration Environment? Well for one a great way to express intent and keep revising that over time (so definitely version control for intent). And of course I want to collaborate on the intent with other humans and with agents. The second part that’s badly needed are better ways of assessing to what degree the system that has been produced meets the intents. Right now, I am doing a lot of screenshotting and then manually feeding that back to the coding agents. A lot of other stuff should happen completely under the hood, such as enforcement of code quality (starting with linting), feeding back error messages, writing and applying test cases, etc. I should be able to look at that if I want to but it really shouldn’t take up any of my attention (and hence screen real estate) unless something is going badly wrong.

In the Beginning was the Command Line is a famous essay by Neal Stephenson, in which he argues against the Disneyfication of computing represented by Graphical User Interfaces (GUIs). And I do have fond memories of programming using nothing but a text editor (VIM) and the command line. Over the years though Integrated Development Environments (IDEs) became more and more popular. And today most programmers are using something like Visual Studio or more recently Cursor.

But IDEs are still code centric and code is going away. Well not away entirely but it will become increasingly hidden just as machine code has been hidden from us for a long time. I grew up at just the right moment in time where machine code and assembly language was still something you learned and occasionally used directly because computers were slow and if you wanted to write a game, core graphics loops had to be hand written in assembly. Since then entire generations of programmers haven’t seen assembly or machine code. It hasn’t gone away entirely but it exists at a layer that is just there and works. Now the same will happen with higher level code. 

Coding agents are doing to source code what compilers did to machine code: push the code below the interaction surface.

This creates an interesting challenge for what the AI native IDE should look like. All one has to do is look at recent iterations of Cursor to see how rapidly this is changing. Cursor started as a fork of Visual Studio (Microsoft open source!) and was immediately recognizable as such. The most recent version on the other hand has an entire section for just dealing with coding agents that looks nothing like a traditional IDE.

The question then is what should the right environment look like? I believe these will be Intent-based Collaboration Environments. In the “nothing is truly new” department it is worth remembering that the idea of intent based programming goes back to at least the 1990s and Charles Simonyi and others even started a company called “Intentional Software” in 2002. They had the right idea but were simply a bit too early. Now we have coding agents that can turn intents into code.

So what would I want to see in an Intent-based Collaboration Environment? Well for one a great way to express intent and keep revising that over time (so definitely version control for intent). And of course I want to collaborate on the intent with other humans and with agents. The second part that’s badly needed are better ways of assessing to what degree the system that has been produced meets the intents. Right now, I am doing a lot of screenshotting and then manually feeding that back to the coding agents. A lot of other stuff should happen completely under the hood, such as enforcement of code quality (starting with linting), feeding back error messages, writing and applying test cases, etc. I should be able to look at that if I want to but it really shouldn’t take up any of my attention (and hence screen real estate) unless something is going badly wrong.