Claude has become very smart, and really good at software development. I'm wanting to experiment with leveraging it as much as possible to increase my own productivity while enabling a (hopefully) nicer experience for contributors of the repos I'm helping maintain.
This means that I'll use Claude to investigate and triage incoming PRs and issues through applying labels, trying to reproduce the issue, or writing entire responses. Ideally, while keeping the same level of quality as if I was doing it by hand without assistance.
I'm guiding Claude, and reviewing what it writes before sending it, but I will still make mistakes (just as Claude will). There's still a human behind this and I'm still accountable for what I write/post, even if Claude is the one wording it.
The goal being that I can be more responsive and active for incoming contributions, while also opening up more time for development work.
I'm also going to be attaching a link to this page at the bottom of some of the things I write so that people who recognize it as AI generated can at least see the reasoning behind it.
- Q: Will this result in low quality AI slop?
- A: I hope not! I'm reviewing everything Claude does, and I'm the one approving its actions of actually posting comments/replies.
- Q: What are the long term plans?
- A: Ideally we can get Claude to be a first line of support, automatically doing some initial triaging in the form of labelling issues, and potentially trying to reproduce issues on its own. For now this is driven somewhat manually.
- Q: I'm not happy about this! Something you wrote was bad and you should feel bad!
- A: That's totally fair. I'm very keen to hear feedback on this, and you can reach me either by responding in issues/PRs here, or maybe an easier way is in the
#python-sdkchannel in the MCP discord (my discord username ismax.anthropicif you want to DM me)
- A: That's totally fair. I'm very keen to hear feedback on this, and you can reach me either by responding in issues/PRs here, or maybe an easier way is in the
Hey @maxisbey thanks for explicitizing your AI usage here. Leaving here one enhancement proposal for your prompts. When referencing code it would be great to link appropriately via a permalink instead of referencing loosely like the example found in this comment when explaining bug 1. Copying the suboptimal code reference down below