Biggest problem for SaaS is that people don't have enough time to be informed buyers. Like in theory you should be able to make a $0.99 annual SaaS app that provides some small service really well and sell it to a million people who need it, but that kind of sale is almost impossible. Most people are either missing out on great software they should be buying, or paying for software they don't need.
Every single AI economic analysis talks about travel planning but none of the AI labs have the primitives (transit routing, geocoding, etc.) in a semantic interface for the models to use.
Hate might be a strong word, but it's natural to feel that way towards a product designed with the primary purpose of increasing LTV per customer per PC.
I think the only restriction that seems problematic is not being able to clone someone’s voice without permission. I think there’s probably a valid case for using it for satire.
You might consider taking this opportunity to travel. You don't need a ton of money to do this, a few thousand dollars goes really far in South America and Southeast Asia. It's also a really exciting time to be building software solo, it's just harder to get paid for it. I never would have been able to bite off the projects I'm working on right now without Claude code. If there's something you want to exist in the world you can really just will it into existence, I assume that will also mean money eventually. TBH it was nice to get the big tech pay a couple years ago, but they weren't as amazing places for actually building software as you might think.
Reply to your comment below this (since hn limits comment depth to 4). The 40ms latency is an average but 90% of queries are getting routed to a single index, latency is worse when the routing goes to 3. Since I already batch the embedding generation I should be able to get hard queries down to like 50ms.
So I don't support queries with a radius larger than 50km (if an AI agent doesn't know where it's looking within 50km there usually is a context issue upstream), but i have a larger h3 index and a tighter h3 index. Then I have a router that tries to find the correct h3 indexes for each query. For some queries I'll need up to 3 searches, but most map to a single search. (sorry I probably won't be able to reply below here since the max hn comment depth is 4)
Mostly usearch since pgvector had some perf problems, though I use postgres during my ingestion stage since postgis is so good. I segment my hnsw indexes by multipe h3 levels.
One thing that’s helped me is creating a bake-off. I’ll do it between Claude and codex. Same prompt but separate environments. They’ll both do their thing and then I’ll score them at the end. I find it helps me because frequently only one of them makes a mistake, or one of them finds an interesting solution. Then once I declare a winner I have scripts to reset the bake-off environments.
Yeah but how often do you finish a AAA game and want to cry at how beautiful it is. You get that feeling pretty often with an indie game. Like something really important is being done by indie devs.
As someone who has been solo developing an app for months concerned ape is such an inspiration. He literally spent 5 years on stardew valley with zero income. It's such a beautiful game and reminds me what you can do when you follow what feels right.
I’m sorry you’re feeling so isolated. It’s really hard to be in a situation where you feel like your attempts don’t go anywhere. But the good news is things can get better really fast. Finding your people in college is one of the best things. I’m not really good at faking a lot of social stuff so I relate a to your frustration. The good news is people respond really well to people who have a genuine interest in them. So it sounds like you need to go find some people you find interesting and interact with them. I was trying to make friends in Seattle recently and I thought that one of the most experienced scuba divers in Seattle wouldn’t want to be my dive partner because they had so much more experience, but my interest in what they were doing closed the gap and now they’re a close friend of mine and I’ve learned so much about scuba diving.
There’s a lot of previously intractable problems that are getting solved with these new embeddings models. I’ve been building a geocoder for the past few months and it’s been remarkable how close to google places I can get with just slightly enriched open street maps plus embedding vectors
I just installed Ubuntu again after a few years, and it’s striking how familiar the pain points are—especially around graphics. If Ubuntu LTS is positioning itself as the standard commercial Linux target, it has to clearly outperform Windows on fundamentals, not just ideology. Linux feels perpetually one breakthrough release away from actually displacing it.
I would love Andrej's take on the fast models we got this year. Gemini 3 flash and Grok 4 fast have no business being as good + cheap + fast as they are. For Andrej's prediction about LLMs communicating with us via a visual interface we're going to need fast models, but I feel like AI twitter/HN has mostly ignored these.
I'm pretty sure xAI exclusively uses Nvidia H100s for Grok inference but I could be wrong. I agree that I don't see why TPUs would necessarily explain latency.