Steve Troughton-SmithI really don't know what development on my own apps looks like, going forward. It's like going back to being Clark Kent when you're moonlighting as Superman. I can't imagine working on my existing apps like I have been with my Codex-built projects, and I'm not sure how to bridge the two ways of working. It's beginning to become difficult to imagine writing code by hand again, and it's only been a month 😐
I think part of the problem is being able to use LLMs with confidence in a greenfield codebase, vs being afraid to break anything in an established app. I am far less likely to ask it to do big things in my apps, and I'm far more likely to back out any changes it does make for fear of the repercussions. There are architectural decisions in my apps that I never would have made had I access to something like Codex at the time, and now I've made them they're too fragile and require too much to fix
Roman@stroughtonsmith A little naïve take but it proves very useful for me: test coverage. The more systems I (Codex) have covered, the more confident I am to change things. I even sometimes want to refactor something and just before that ask Codex to generate snapshot tests for many screens. To make sure not a thing looks even slightly different after it’s refactoring
Rhys Morgan@stroughtonsmith See, this is where I feel AI really shines. It’s allowing me to pay down YEARS of tech debt at work.A lot of it has decent test coverage, or I can figure out how something works (e.g. it’s using Promises and I want to migrate not just to Combine, but modern Concurrency). But using plan mode, using it to build my own understanding, etc. is helping me with those larger refactors that “require too much”.
Paul Haddad 
@stroughtonsmith it's really good for clearing out/changing old complex code. I feel far more confident in changing something low level that was written years ago when I can have a 2nd set of eyes looking over my changes to make sure I didn't miss anything.
Taylor Hadden@stroughtonsmith What kind of architecture decisions are you thinking about? I’m really curious about any examples.
Chris Clarke@stroughtonsmith I'm really curious how you envision approachiing the maintenance phase of an app developed like this. I can see how one would approach new features, but what about bugs? Do you trust the LLM to "fix" code without changing something else, do you need to break into the code yourself?
@checlarke I think people overthink this; if you, as a product manager, had to fix bugs in your team's software project, would you get stuck in yourself, or would you delegate? That answer will vary person to person, but I think it's informative here.
Most likely, you'll delegate and trust your team — you assign a model to fix it, and if that model doesn't work, you might try another, and if that doesn't work you might try education or changing processes to make issues like this less common
@stroughtonsmith thanks - I need to think about this. The thing is the capabilities of these models are changing so fast that my own mental model is having trouble keeping up.
Work Less@stroughtonsmith Inspired by you, "i" (Claude) rewrote an entire community site built in (very poor) Classic ASP, VBscript, using a mix of SQL and Access databases... to C# and blazor, cleaned up the database schema and consolidated it on SQL, then containerized it under linux docker images, over the course of two weeks. Shipping to production almost daily as i went.
I am now able to retire a Windows VM i've been running for years, that replaced shared hosting i paid for, for years. The users never even noticed a difference, it still has 1998-style HTML views to maintain the vibes. Literally 25 years of "I need to rewrite this," procrastination solved in two weeks.
Just do it.
Benjamin Smith@stroughtonsmith By my experience with AI generating code for me, what I find:
1) Only give it small pieces, maybe 200 lines of code
2) In a very narrowly defined niche
3) that you check out immediately afterward
4) You do have unit tests, right?
I find AI to be VERY useful when beginning layout. I have a problem, I'm trying to solve, here are the variables I have to keep in mind, here are the concerns I know about, what do you think Chat/Gemini/Claude? I make it convince me.