Home

Ordering Chinese but this time with AI, or the Merit of Bad Ideas

Author: Andrew Gallagher

Published at: Jan 20 2025

Back around 2012-ish, I was a member of a medium-large Facebook group named simply "ideas". In the group, posters wrote single sentence concepts and left them for other group members to consider. Ideas ranged from "a car powered by bong water and positive vibes" to "two Chief Keefs". You get the picture — just a bunch of guys fucking around. It was a great time.

In software we strive for a mostly logical, causal relationship between what we're developing and what we're trying to solve for. What I mean is that we go out and talk to people, identify some unaddressed need, figure out if the market is viable, develop some kind of widget and then iterate against that until you can't iterate any more. That type of thing. Developers are conditioned into this structured way of thinking — every decision and LOC should be driven by some fundamental business or technical cause.

And this is a good thing! Not a particularly profound assessment on my end. We should be working for some practical end most of the time — electricians, lawyers, and veterinarians all need software in the modern world and software developers need to put food on the table. There is also much beauty and purpose to be found in the world of utilitarian things, even in software. But I bring up that earlier point about the joy in having easy, half-baked ideas that go nowhere for a reason: I think, sometimes, simple bad ideas are good.1

Again, this is unfortunately not a particularly profound assessment on my end. UX people used to talk about "ideation" all the time, and about how the best way to have a good idea is to have a high volume of ideas while accepting that most of them will be bad.2 I think this is fair, but I also want to more specifically point two killer features of a bad idea:

  1. Bad ideas are really mentally inexpensive to make. Just think of the first, worst thing and roll with it.
  2. A bad idea has a built-in timebomb. Working on a bad idea has the opposite of the feedback loop associated with a good idea — you will receive no positive feedback, everyone — including yourself — will question why you would spend 5 more seconds on this idea, and a crushing boredom will inevitably box you out. In the world of bad ideas, you have to get in, hold your breath while you do your thing, and get out. If you’re lucky, you learn something.

Since my last startup wound down, I've been looking for a good job :-) and it has been pretty easy this go-around trying to figure out what it is that the market values right now in the valley. As a product developer, I haven't had much exposure to working with newer AI services and wanted to tread into the murky waters.

When you're exploring a new technical domain, it can be tricky to figure out what to build beyond the TODO app, because, well, you have no idea what any of this technology can even do.3 You get where this is going — let's roll with our first, worst idea. What am I eating tonight? Chongqing Numbing Spicy Noodles. How am I gonna get it? You already know what it is playboy — we're going to slam together 9 different AI services4 for $14 of pickup.

Now I mean, in 2025 taking something that's been done to absolute death (food delivery) but also Now With AI™️ is borderline illegal or should be. But when exploring a new domain, being forced to think quickly, take as many shortcuts as possible, and get something barely working, the confines of an ill-fated venture lend you so much speed and structure. Enums causing some weird Prisma error on your client? Fuck it, covert them to strings. Taking a bit too much time to flesh out a service that can handle different providers? Just write that shit in the route handler. What's the worst that can happen? You don't have to pay down tech debt if you just blow the thing up.

And after about 3 days, I was at my absolute limit. I could not bear the thought of putting even one more keystroke towards this repo. But that's good! That's the built-in shutoff kicking in. This was a bad idea, so walking away has always been the smartest thing I could actually do. The repo, without a single fork or star, will unceremoniously decay in cyberspace. It is, at best, a cairn for Python AI/ML devs trying to figure out how to structure a TypeScript application repo. At its worst, it's still fodder for the very machines that power it.

  1. Note that it's important to not conflate "bad ideas" with "side projects", "unproven theories", or just "having ideas". For our purposes, a "bad idea" is a specific variety of side project that is not highly aligned with a developer's personal interests, is not of notable technical interest, and will not make you any sweet cash. You know this all ahead of time. It should feel stupid 100% of the time.
  2. I am fairly certain that UX as a field is dead post-ZIRP, and has been folded into design. RIP. It is what it is. Pour one out for Don Norman. That said, I'm relieved I hear about their concept of "ideation" less and less, because the phrase is otherwise like, only associated with suicidal ideation for me. Listening to these guys, it literally just sounded like they all just wanted to kill themselves all the time.
  3. I am aware that "reading about things" or "picking up a goddamn book" is always on the table, but I personally believe there is a real threshold where reading becomes less effective than just punching keys.
  4. I do want to actually be sincere/serious here and note that I actually met with Jordan, co-founder of Vapi and he is looking for good engineers. If you want to work incredibly hard and probably make a LOT of money, you should reach out to him.