I meant to say that would can give each model the same task and keep the code that works best for the long term. Or, you can merge the best parts of each models changes.
You can give the same model the same task multiple times and it often results in different edits made. That is just how AI works
I would think that since both Codex CLI and Claude Code are currently so cheap for the value they provide, you could keep both. Have Claude and GPT-5 work on same tasks simultaneously in separate worktrees or have one review the changes the other makes. All for for less than $500/month.
Claude Code is a huge step forward IMO.
Nah. I bought the Atari 2600 in 1977 which led to me buying an Apple II with an amazing 48k of RAM in 1978 and upgrading that from just cassette player input and output to Woz’s amazing 5.25” floppy drive with 110KB of storage for my programs and data.
Computing has advanced every year since.
You can always buy a second Max plan so when or if you hit a limit, you won’t be left waiting a few days to continue building. $300 a month is a very small price to pay for Claude.
Ask Claude to explain how to use Claude.md to give Claude the context it needs. /init will get you started in a project. You can easily add to the context with ‘#’ or just edit the file directly.
Mastering Claude.md is essential to get the most out of Claude Code. This is the best mechanism to give Claude Code the proper context. You can have many Claude.md files either globally to give context for all your projects, local to a specific project, or in a specific folders in your any project.
Why did you specify “think harder” instead of “ultrathink”?
Just curious…
So, I don’t think you can set an agent thinking mode using just slash commands as one slash command is one prompt sent to the LLM. You would need to use a Hook to inject the ultrathink keyword into selected prompts.
I’m just getting started with Claude Code so haven’t implemented any hooks yet.
Note the keywords of “think harder” and “ultrathink” aren’t really processed by the LLM but simply set the number of max tokens the LLM may use when processing the current prompt. This is the way Anthropic tries to save on inference costs for those users that are on the Pro or Max plans.
You can quickly make a Command in Claude Code to do this exact thing. This is very easy. You should also read up on adding Hooks into Claude Code.
Between commands, hooks, and MCP, you have full control to customize CC to optimize your dev workflow as you work.
Musk drunk with power.
