Feature Request: Building on Architect Mode with an /agent Command

[swushi]

Issue

Description:

First, I want to acknowledge how valuable architect mode is. The two-step approach with an architect and editor model produces excellent results and shows aider's commitment to quality outputs. It's a perfect example of how thoughtfully designed aider is.

However, architect mode still operates within a fixed pattern - an initial planning step followed by a single editing step. What I'm proposing would build on this foundation to enable more flexible, iterative workflows.

Proposed Solution:

A /agent command that would allow for customizable multi-step processes beyond the current architect/editor pattern. This could support:

• Iterative cycles of plan-code-test-refine
• Self-validation steps where the agent checks its own work
• Specialized workflows for debugging, documentation, etc.

Like architect mode, this would maintain aider's clean user experience but extend its capabilities for complex problems.

Why This Matters:

With tools like claude code demonstrating the effectiveness of more autonomous coding approaches, I believe aider should evolve in this direction while preserving what makes it great. The coding landscape is shifting toward tools that can handle more of the cognitive load, especially for repetitive tasks.

The beauty of implementing this as a command is that it keeps aider's core simplicity intact for users who prefer the current approach, while opening new possibilities for those who want more automation.

Would y'all consider extending architect mode's success with more flexible agentic capabilities?

Version and model info

No response

[swushi]

#3362 - Similar issue with a decent amount of traffic.

[Dima-369]

I like how Cline does it. You can start in Plan mode, ask questions as many times as you need, refine the LLM ideas on how it should be implemented and once you are happy, you switch to Act mode and let it implement that way.