I've tested these new AI code agents:
🤖 Claude Code
🤖 OpenAI Codex
🤖 Google Jules
I wanted to make changes to the codebase of my Salesforce Flow Extension at https://lnkd.in/gbsv2DqN
The extension has reached 2k weekly users recently and I intend to release a new version compatible with the most recent models. So I've tasked these 3 AI tools to look at the code and make suggestions.
Claude Code and OpenAI Codex are command line interfaces with the ability to read your local drive and modify the code as per your instructions ("refactor this", "optimize such and such", etc.).
🌟 🏆 Claude Code - shown in the 1st screenshot - was really good. It showed me the planned changes before applying them and most of the time they were good changes. It can also read your code and make a list of suggestions.
🚮 🤮 OpenAI Codex - shown in the 2nd screenshot - was terrible. It churned a lot without presenting coherent outputs and then gave up, apologizing. I had to restore a backup of the code to clean up that mess.
👍 Google Jules - shown in the 3rd screenshot - worked decently, but not at the level of Claude Code. It works a bit differently, operating on the repo directly rather than in your local drive. It also hesitates more, asking for multiple confirmations before committing. It is a free option and good enough.
