A Second Opinion From the Other Guys

Same directory, same question, different model. In the last lesson I had Claude Code review Moebius, my Elixir data access library. Now I open the project in Codex, OpenAI's coding app (which, like the Claude desktop app, is a little confusing about when you're chatting and when you're coding), pick Astra, their frontier model, crank the effort to extra high, and dictate the review again.

This is an Elixir project that I recently upgraded from an older version of Elixir. It's about two years old. I upgraded it. I rolled into a new driver away from Postgrex. What I want to know is, is the code safe? Is it idiomatic Elixir? It's an open source project, so I need it to be as safe as possible, without SQL injection, also no lockups. I don't want dirty reads or dirty writes. I want to make sure everything happens in a transaction, and that the code is idiomatic. Bottom line is, since it's open source, I don't want to be made a fool of, for lack of better words. Take a look at this code and give me a review. Tell me what you think.

Astra found things Claude didn't lead with. Connection handling gets tricky in Elixir because you're in a functional world with side effects on top of the actor model, and sure enough, a rejected parameter could cause trouble. It also caught that a full text search silently removes existing filters, which I had never considered. SQL injection, transactions and concurrent writes all looked sound. There are legitimate P1 and P2 fixes in there, and I'll make them and push 5.0.1.

The point is not to tear Moebius apart or to crown a winner. Each model has its own take, because each harness has its own system prompt and its own way of thinking about code. People who do this for real bounce between both: have Astra check the code and Opus write it, or the other way around, whatever works for you. The tooling dictates everything, and we'll dig into exactly how in a later chapter.