This paper is basically a vision / collection of ideas of how to further integrate agentic software engineering into a software development process:

One of their premises is that AI-enhanced IDEs are mostly focused on code and less on actual sound process. I'd agree that the typical implementation of orchestration over architecture, analysis, coding and debugging is a rather poor implementation of a software development process, especially because it's a very primitive LLM replaces human concept where human roles or function titles are attempted to be replaced by prompts.

If you've ever worked in a large product development organization, you know that this is not the essence of the process of coming from a requirement to a successful delivery. The success often doesn't come from the roles you have defined, but from the optimization of the flow and the removal of friction on interfaces (between people and automation and among people). This paper seems to recognize that.