Session 1
Iterate in the chat — and undo anything
Goal. Refine the script conversationally, then learn the single command that makes experimenting risk-free.
▶ Try this prompt
Also write the results to a summary.csv file, and skip any columns that are completely empty.
Send this as a follow-up in the same Aider session — it keeps editing the existing file instead of starting over.
- 1Keep the session open and describe your next change in plain English. Aider applies it to the existing code and commits the new version — each instruction is its own clean Git commit.
- 2If a change is wrong or you don't like it, type
/undo. Because Aider committed the edit,/undocleanly reverts that last AI commit — nothing is silently overwritten. - 3Work in small, focused steps: one clear request at a time gives the model less to misread and keeps each commit easy to review.
You'll see. The script updating in place across several requests, each one a separate commit — and a clean rollback whenever you run /undo.
Cost. Each request is one metered call to your chosen LLM; grouping related changes into one clear message spends fewer tokens. /undo is free — it is just a Git revert.
Takeaway. You drive Aider conversationally, and because every edit is a Git commit, /undo makes every experiment safe to try.