Steve Troughton-SmithFeb 24

RE: https://mastodon.social/@kocienda/116126204434018823

I think having a master reference copy of my apps, built mostly by hand (in my case for iOS/Mac) is still important.

But the two Android apps I shipped last week, and two Windows apps I'm working on now, I effectively haven't written (or even reviewed) a line of code. If the reference (i.e. a detailed plan) is good enough, and if you test it well enough, that's kinda all you need.

Your source code really isn't precious, and if you're that much of a perfectionist you're probably not shipping

Show threadCallum JonesFeb 24

@stroughtonsmith (I haven't used and don't plan to use Codex or similar):

To me this argument reads like your software architecture is irrelevant to you, almost the "AI is the compiler" ideas I keep seeing.

If you aren't reading the code, how do you understand the maintainability of it, and what do you do when the inevitable bugs occur? I find debugging my own code much easier than someone else's, never mind something written by something with no perspective/logic/reasoning.

Show threadSteve Troughton-SmithFeb 24
@seejy it maintains the code, it fixes the bugs? If it can write 1,000 lines of code in a few seconds based on a few-word prompt, why would I need to get in there myself except for the most critical of emergencies. It has your entire codebase and chat history as perspective, which is certainly more than I can keep in my head at any one time
Show threadMartin
@stroughtonsmith @seejy I have high respect for your work, but the examples you build are such low-risk things, aren't they? A color palette app. An offline game.
What about authentication and managing user data? Or payments? Would you still say "why bother going in there" until the critical emergency occurred? Let's say it even gets 95% correct, what about the 5% where user data is leaked. The undefined edge cases where payments fail?
Sounds like a nightmare to me. 🦇
Feb 24 at 4:16pmWeb
Show threadSteve Troughton-SmithFeb 24

@mrtnlst @seejy that seems so obvious I didn't bother bringing it up — of course my stuff is low-risk, so is most software. If security is your utmost concern, don't use any of this stuff. If you're in any way security-conscious, you already know this. If you need a security expert, hire a security expert.

Most people don't, and most people already have software with plenty of security issues, written by hand, that they're oblivious to.

That's not an argument to avoid this stuff entirely, imo

Show threadMartinFeb 24
@stroughtonsmith glad to hear that though! Thanks!
Show threadCallum JonesFeb 24

@stroughtonsmith aside from the already well documented downsides to AI, I am not sure how I would feel about putting my name to software created through AI.

For one, I like solving technical problems, and I like writing code. I'm not sure I'd get the same satisfaction from prompt engineering (though again, I have not tried it).

Is it just that it is a means to an end in order to get e.g. the Windows apps? Do you feel an attachment to this code too?

Show threadSteve Troughton-SmithFeb 24
@seejy it's just a different abstraction level. SwiftUI is a different abstraction level from UIKit/AppKit, masks a ton of complexity, and introduces a ton of its own bugs in the process, but people love it. You're still solving problems, but problems at a different abstraction level. I feel great about what I've been building on Android and Windows, I've got to lean into all the things other developers tend not to, and express my taste. It's very much a skill, different part of the brain though
Show threadSteve Troughton-SmithFeb 24
@seejy I love what I've built — the underlying code could be swapped out underneath in an instant and not matter much to me (and often is as something like Codex refactors things based on new prompts), as long as the finished product looks and feels how I intend it to
Show threadSteve Troughton-SmithFeb 24
@seejy (and these are interesting questions, because a few weeks ago I would /never have considered/ building anything like this was viable. But now I'm several shipping apps later, and have got to bring light to so many of the ideas on my shelf)