My favorite thing about using two AI coding assistants? When they disagree. š
byā¢
Iām building a digital marketplace without a traditional coding background, and one of the most interesting things Iāve experimented with is connecting Claude Code and Codex so they can collaborate.
Sometimes I ask Claude to handle a task from start to finish and have Codex review the implementation afterward.
Other times, I ask them to work together throughout the process.
And then I get messages like:
āCodex gave us a NO-GO. And heās right. He found two issues I missed.ā
š
I actually love seeing that.
Because when youāre learning to build with AI, one of the biggest challenges is knowing whether a solution is genuinely good or just sounds convincing.
Of course, two AI assistants agreeing doesnāt guarantee the code is correct. I still have to test things and question the results.
But watching them challenge each otherās decisions has completely changed how I approach development.
Iām curious: If youāre building with multiple AI coding tools, do you give them different roles, or do you let them collaborate on the same task?
And have you ever had one AI catch a mistake the other completely missed?
19 views
Replies
Back in the days we used to print out thousands of lines of Perl on dot matrix paper just to manually track memory leaks over the weekend. Now you just start up two language models and let them debate about state management while you make a cup of coffee.
One agent generated a report with a confident performance number, and the second agent wrote tests asserting it. It took a full day of debugging to realize nobody had measured a single thing. It was just a placeholder hardcoded into a mock.
Give the reviewer the goal, diff, and tests, but not the implementerās explanation. Have it reproduce important claims; two agents can agree on the same bad assumption.