Well, the crazy thing I'm working on (100% self-funded thus far): https://trivyn.io. The main idea is moving most of the reasoning to the symbolic layer so the "neuro" piece can be a small model able to be self-hosted on reasonable hardware.
True. Sometimes I'll run front-end and backend work in two different claude instances, but always on the same project/product. I'll have "reviewer" instances in opencode using a different (non-Claude) model doing reviews, that's about as much as I can handle. You've got to supervise it while it works. I do have to stop claude from time to time when I catch it doing something naive or unnecessarily complex.
Yea, my issue with Opus 4.5 is it's the first model that's good enough that I'm starting to feel myself slip into laziness. I catch myself reviewing its output less rigorously than I had with previous AI coding assistants.
As a side project / experiment, I designed a language spec and am using (mostly) Opus 4.5 to write a transpiler (language transpiles to C) for it. Parser was no problem (I used s-expressions for a reason). The type checker and transpiler itself have been a slog - I think I'm finding the limits of Opus :D. It particularly struggles with multi-module support. Though, some of this is probably mistakes made by me while playing architect and iterating with Claude - I haven't written a compiler since my senior year compiler design course 20+ years ago. Someone who does this for a living would probably have an easier time of it.
But for the CRUD stuff my day job has me doing? Pffttt... it's great.
Trivyn: Ontology-first knowledge platform. Runs on a single machine, via a single executable. I wanted a simpler alternative to the large complicated enterprise products that tend to dominate this space.
I'm really trying to get a private beta out the door by Christmas. I do plan to have a free version for academic/personal use.
Backend is written in Rust, uses oxigraph for its triple store.
I've found most LLMs I've tried generate better code in typed, procedural languages than they do in something like Clojure.
From the perspective of a primarily backend dev who knows just enough React/ts to be dangerous, Claude is generating pretty decent frontend code, letting me spend more time on the Rust backend of my current side project.
> generate better code in typed, procedural languages
Better in what sense? I've been using Anthropic models to write in different Lisps - Fennel, Clojure, Emacs Lisp, and they do perform a decent job. I can't always blindly copy-and-paste generated code, but I wouldn't do that with any PL.
Yea, my house is on Starlink... so at the moment I'm working in the back room of my wife's restaurant (she has cable internet). At least I get free coffee here.
