AI killed Agile
(Why sprints and scrums no longer make sense when AI can write the code in hours)
Watch the YouTube video at: https://youtu.be/t94DyXLh_u8
AI has changed Agile forever. We never used sprints because we loved sprints. We used them because they were faster than the alternative. Now that an AI coding agent can implement an entire application in hours, the reason for Agile is gone.
I have led many development teams over the years, and this is the biggest change to how we build software that I have seen.
The Reason We Used Agile Was For Speed
The teams I led never adopted everything about Agile. What we did adopt was sprints. We produced a workable application in small pieces so the end users could give us feedback sooner, rather than building the whole application at once and finding out at the end that we got it wrong.
The reason we did this was speed. Writing code was slow and expensive, so you wanted to find your mistakes after two weeks of coding, not after two years.
That was the entire bargain. Agile was not the goal. Getting to the right application faster was the goal.
What Is Waterfall And Why Did It Fall Out Of Style
Waterfall is the approach where you do everything in order, one time: gather all the requirements, design the entire application, build all of it, test all of it, then deliver it.
There was never anything wrong with designing the whole application up front. The problem was the build step. It took months, and sometimes years. By the time the end users finally saw the application, the requirements had changed, or they had been misunderstood from the start, and fixing that meant more months of coding.
Waterfall did not fall out of style because the design was bad. It fell out of style because the coding was slow.
What Is The New Agentic (vs. Vibe Coding) Development Process
There are two ways to build an application with AI, and they are not the same thing.
Vibe coding is working with the AI feature by feature, much like the Agile method. You ask for a feature, look at it, then ask for the next one. This turns out to cause a lot of AI drift. The AI breaks existing code as it adds new code because, just like a human, it is not holding the entire application in its head.
Agentic coding avoids this by designing the application all at once, in Markdown (.md) files. The specifications can run to thousands, sometimes tens of thousands, of lines across several files. Then the AI implements the whole design in one pass.
Yes, this is much like waterfall. The difference is that the step that took months or years now takes hours.
The code the AI creates does have bugs, so the process is not finished when the code is written:
- Tell the AI to actually run the application, click every button, and test every feature.
- Tell the AI to perform a security review where it tries to hack into the running application.
- Let the AI iterate, sometimes for hours, testing and fixing its own bugs.
- Review what the AI reports back. Humans still have to make certain decisions, such as resolving conflicts in the design.
Only then is the application ready for the end users to test and use. This is why the process takes two to three days rather than an afternoon, even for very large applications.
The New AI Agentic Process Moves As Fast As Humans Can Move
The part that has not changed is the people. You still have to gather requirements. You still have to show mockups to the end users and refine the application based on their feedback. For a large application this still takes weeks, and sometimes months. Coordinating meetings, and allowing people time to discuss and research, takes as long as it ever did.
What has changed is that the humans are now the only thing setting the pace.
Mockups are the best example. Tools like Claude Design and Google's Stitch can create mockups directly from the .md specification files. I have found this is the best way to work with the humans when designing an application:
- The end users react to screens, not to thousands of lines of specifications.
- Their feedback goes back into the .md files, and new mockups are created in minutes.
- The mockups and the specifications never disagree, because one is created from the other.
When everyone signs off on the mockups, the design is done, and the application is a few days away.
So Where Does This Leave Agile?
| Waterfall | Agile | Agentic | |
|---|---|---|---|
| Design | All at once | Sprint by sprint | All at once, in .md files |
| Coding | Months to years | Two-week sprints | Hours |
| Testing | At the end | Every sprint | AI tests and fixes for 2 to 3 days |
| User feedback | After delivery | After each sprint | On mockups, before any code |
| Slowest part | Coding | Coding | Humans |
Sprints and scrums existed to manage the slow, expensive part of software development, which was writing the code. That part is no longer slow or expensive.
What Agile got right survives: get feedback from the end users early and often. We just get that feedback on mockups now, before any code exists, instead of on working software after each sprint.
AI did not bring back waterfall. It removed the one thing that was wrong with it.
Learn More
You can learn more in my new book, Agentic Blazor: How I Build Production-Ready .NET Applications in Days.

