I've shared the full transcript of every agentic coding session from implementing the unobtrusive Ghostty updates and provided commentary alongside about my thinking and process.

I'm regularly asked to share non-trivial examples of how I use AI and agentic coding tools and this felt like a golden opportunity to walk through my process with a well-scoped, non-trivial, real-world, shipping feature.

Total cost: $15.98 over 16 sessions.

"Vibing a Non-Trivial Ghostty Feature" https://mitchellh.com/writing/non-trivial-vibing

Vibing a Non-Trivial Ghostty Feature

Mitchell Hashimoto
@mitchellh some really good points here. Thanks for sharing.

@mitchellh hey Mitchell, thanks for this post, it's a really interesting window into this stuff.

I know this might be an annoyingly common request but have you spoken to the ethics of all this on record anywhere? I want to love Ghostty, but I'm more and more uncomfortable with LLMs and the ethics surrounding them.

@mdiluz I haven’t, and obviously Mastodon won’t be a good place to do that. Note that Ghostty itself has no AI features. If software written with AI assistance is enough by association to make you uncomfortable then I’ve got bad news for you for pretty much the entire software ecosystem…
@mdiluz more directly though: I have thoughts. One of my closest friends is chief nuclear officer of a nuclear energy company and another who I spent a few hours with yesterday is a very successful IP lawyer dealing with a lot of AI stuff. We all talk about it constantly. I have thoughts. But not really ready to put em down
@mitchellh yeah no worries, it's such a complicated issue for us as developers - especially if wanting to use "handmade" software when possible. You're 100% right that it's a struggle! I appreciate the reply, and your transparency and intellectual curiosity, it's at least a step in a different direction to others I see on my feed.
@mdiluz expect him to be fine with them. thats why he is using them, posting about them, and sharing about his process of using them.

He just doesnt want to upset people who don’t like them
@mitchellh I respect that you're so open about it, but it does mean that your product is something I will absolutely never engage with because of it.
@nullpotential all good, I have no problem with consumer choice :)
@mitchellh @pjacock thanks for writing this up.
@mitchellh Great post, thanks! I am curious, as you work with the LLM, how do you manage with the idea it may trash existing in process code? I usually "checkpoint" by adding code that is on the right track into the git staging area, but only commit after I have reviewed and cleaned up manually. But I wonder how others do this and I don't see it talked about often.
@emil I "jj new" aggressively, sometimes with chains of up to like 10 unnamed commits. Then I "jj abandon" or squash things around.

@mitchellh I finally had an opportunity to do some agentic coding for the first time myself this weekend and was pleasantly surprised by the results.

I had a small problem to solve and asked Copilot to do a simple file import for me to get started because I never remember how to do it. That step went so well I ended up spending the day having Copilot write the whole utility for me, including refactoring, tests, and docs.

Thanks for sharing. Nice to compare my own experience with yours!

@mitchellh Thoughts on J blows recent tweet on this (semi-related). Screenshotted to not direct traffic to X
@treydogsacid I think this aligns with what I wrote in maybe a slightly more inflammatory way. I guess he’d agree with me
@mitchellh loved (and very much related to) the coda about making breakfast for the family whilst vibing. In my attempts to work in this style (less on code, more on RFCs and such), I find the freedom to context switch to be a double-edged sword — more time to do other tasks, but I also rarely find myself in anything like a flow state.