Don't understand the semantics of architect mode

[RuminationScape]

Issue

Several aspects of architect are not clear to me:

1. What happens between the 1st step and 2nd step of architect prompt? Is the output of the 1st step fed into the 2nd step for it to implement the code?
2. If the 1st step provides some code implementation, is that ignored at the 2nd step? Using Gemini + Sonnet combination, if Gemini provides the code in the first step, the 2nd step Sonnet does nothing and no code is added. For example, Sonnet in the 2nd step would respond with the following without any code changes:

"Thank you for the detailed overview and implementation steps. I don't have any specific changes to suggest at this point. The code you provided looks well-structured and follows best practices. I'm ready to proceed with the next steps whenever you're ready to share them."

From these issues, it seems that in architect mode, the 1st step should be a request for feature implementation strategy without any code. The resulting output is then an instruction guide for the 2nd step to implement. Is that correct?

Version and model info

Aider v0.72.3
Model: gemini/gemini-2.0-flash-exp with architect edit format
Editor model: openrouter/anthropic/claude-3-sonnet with whole edit format
Git repo: .git with 10 files
Repo-map: using 4096 tokens, auto refresh

[mmorrison24]

I've had a similar issue - namely - how is architect mode different than /ask mode?

With the /ask workflow: I've never had an issue with the "step 2" referencing the /ask step.
I have also not had any issues using /ask multiple times and maintaining flow

But your issue begs the question on how this is meant to work

[RuminationScape]

Probably the difference between /ask flow and /architect flow is that /architect allows the use of 2 different models within the same session, whereas the /ask + /code flow only uses 1 model.

This assumption would be falsified if --editor-model parameter is used for /code operation.