I definitely find value when working with models to run things by for another perspective. My favourite moment was when I ran a plan that Replit made by Claude, Claude had some opinions about it, brought them back to Replit, and Replit's response was "Claude is over-reacting", made me chuckle. I felt like such a middle-man.
Delegation heuristics are prob the tricky bit. Codebase investigator is obviously a win, but routing decisions get interesting when you're weighing cost against latency for different task types. Multi-model setups will just be how this works once people figure out the right handoff rules.
what i have found to work well is to use it during the planning phase especially when working on an existing project e.g; if i want an adversarial review of the architecture.
Interesting. I have not used codebaseinvestigator before but as the maintainer for a small-medium sized codebase (~150K lines of code) I think this is something I should look into. Thanks for the tip!
This is great! I find myself using Gemini to double check Claude, idea being that they are different models so can be force multipliers/sanity checks for one another
I definitely find value when working with models to run things by for another perspective. My favourite moment was when I ran a plan that Replit made by Claude, Claude had some opinions about it, brought them back to Replit, and Replit's response was "Claude is over-reacting", made me chuckle. I felt like such a middle-man.
Delegation heuristics are prob the tricky bit. Codebase investigator is obviously a win, but routing decisions get interesting when you're weighing cost against latency for different task types. Multi-model setups will just be how this works once people figure out the right handoff rules.
what i have found to work well is to use it during the planning phase especially when working on an existing project e.g; if i want an adversarial review of the architecture.
Interesting. I have not used codebaseinvestigator before but as the maintainer for a small-medium sized codebase (~150K lines of code) I think this is something I should look into. Thanks for the tip!
You are welcome! please do leave feedback, will be happy to accommodate.
This is great! I find myself using Gemini to double check Claude, idea being that they are different models so can be force multipliers/sanity checks for one another
thank you so much for the feedback! if you find any issues with the plugin, please let me know.
[dead]