AI-powered coding tools like Cursor, GitHub Copilot, Crystal, and Claude Code are shortening the development cycle which makes planning and process delays more salient as the ratio of planning to development increases. Feature Team work needs to adapt.
More rapid development breaks the waterfall process
In traditional waterfall, before AI, you would spend 2 to 10x more time on development that than in planning. This made it feasible to have long, process-oriented planning cycles.
This ratio changes with AI coding where for every hour of planning you might only spend a single hour on development moving the process bottleneck from engineering to planning. To address this, Andrew Ng recently proposed a 1 PM to 1 Engineer or 1 to 0.5 ratio.
More rapid development has implications for Agile as well
The point of Agile is to break the problem apart in order to:
Before AI, Agile development looked like this:
As development cycles are shortened by AI, agile development looks more like this:
I've been interviewing many engineering and PM leaders about this dynamic. Some have said their AI-first teams are following standard process and are frustrated. Others have thrown process to the wind and are following a process like this:
Neither is right. To work effectively with AI as a feature team, there needs to be a lighter-weight, faster process.
AI coding platforms are enabling PMs and UX Designers to prototype full experiences more quickly. Tools like Lovable, Bolt, and v0 let PMs get from idea to customer feedback as quickly as possible.
An PM can help initiate an AI-speed appropriate process with two things:
Its not clear that these PM mockups / prototypes are helpful as code for the developers. So far, for Stravu, they have not been. But, they are helpful as input into the Feature definition.
A more lightweight, but still coordinated/managed process for a feature team working, prototyping, and coding with AI looks like this. This is how we are trying to do it at Stravu:
At the center of this process is a unified Plan that contains the requirements, prototypes, architecture, technical approach, and to dos for the Feature. These words are paired with the code that is being developed. The plan and the code are kept updated and in synch as the team iterates rapidly. This approach enables coordination, rapid iteration, change based on everyone's feedback and learning, and everything staying in synch
In general, AI code increases speed but decreases quality. Some organizations are willing to take this tradeoff. Some are not. Many spend some of their saved developer time fixing the issues in AI's code. The following 4 inputs help to improve the quality of AI Code's output but do not mitigate it entirely:
We are all figuring this out as it happens. What changes has your team made to adapt to AI-powered development?
Stravu provides a way for feature teams to work together on unified, collaborative plans as they iterate and develop with AI.
If you are interested in exploring this with us, sign up for our Beta today!