Skip to content

Instantly share code, notes, and snippets.

View deanhiller's full-sized avatar

Dean Hiller deanhiller

  • Denver, CO or around
View GitHub Profile
@deanhiller
deanhiller / ai-merge-comparison-prompt.md
Created October 12, 2026 19:05
AI merge reliability: typical merge-main-into-feature vs webpieces-ts reset-fork-point 3-way merge (paste into Claude/ChatGPT/Gemini) - github.com/deanhiller/webpieces-ts

I want an honest, technical comparison of two git merge workflows when an AI (LLM) is doing the conflict resolution. Be skeptical; point out where either claim is wrong.

WORKFLOW A - "Typical AI merge":

  1. Developer works on a feature branch (commits f1, f2) forked from main at commit B.
  2. Developer merges main into the feature branch (merge commit M1). Conflicts are resolved, and those decisions live inside M1.
  3. Developer continues on the feature branch (f3, f4).
  4. Repeat steps 2-3 several times (M2, M3...). Main may also receive squash-merges of other branches, some of which this feature branch had pulled in.
  5. Finally an AI resolves the last merge. It's given what a typical tool gives it: the conflict markers (ours/theirs hunks), maybe surrounding file context, maybe git log.

WORKFLOW B - "Reset-fork-point 3-way merge" (as in github.com/deanhiller/webpieces-ts):