When I show people how I work with agents, I’ll often hear, “Wait, isn’t this just waterfall?” And the answer is: well, kind of.
- Plan out your work by sitting down and talking to the agent
- Work with skills that help fill in the gaps before any code gets written
- Keep going until you and the agent are confident the plan is ready to be built
- Have your agent build the feature
- Test the output
- Iterate, iterate, iterate
The reason people think this feels like waterfall is that my planning phase is often as long as, or even longer than, the code-writing process. I may spend 2–3 hours planning, then have an agent spend 20–30 minutes writing the code, followed by another 20–30 minutes of QA and iteration.
The problem with waterfall was never that planning is bad. The problem was that you could spend three months designing everything upfront before you even had the chance to find out whether your design was wrong. Agile sought to fix this by shipping and iterating quickly, bringing user feedback into the mix before those decisions became too expensive to reverse.
Three months spent designing the wrong product can easily turn into another six months undoing it. And if you still don’t listen to users the next time you try, there’s no guarantee you’ll get it right the second time around. But now, thanks to agentic development, the cost of reversing course can be incredibly low.
If I spend half an afternoon building a feature that doesn’t resonate, I can reverse course before the day is even done. Sure, I’ll never get those three or four hours back, but I may still have time to try out a second, third, or fourth version and see if there’s a better approach.
AI doesn’t inherently make every idea you have better, and it doesn’t remove the need for user feedback. What it does is dramatically lower the cost of being wrong and help you iterate toward the right answer much faster. I don’t know what to call this approach yet, but planning for hours, building in minutes, and trying again in the same afternoon is a completely new way of working. Calling it waterfall misses the point.