It's been a great multiplier for me in similar ways. The "dreamiest" thing has been that it has freed up time that I would normally have spent doing sprint work, to work on things that just don't make the cut until it's bad enough to deprioritize other work.
Over the last few months, I've been digging into performance problems with a high throughput service that my team owns. I started working on the problems in my own time, put out short and medium term improvements that legitimately avoided operational issues, and started developing an alternate architecture that should meaningfully address the problems for the long term.
I've learned new things and made improvements that probably wouldn't have ever gone in otherwise.
This really resonates with me, and I'm only a decade and change into my career. I use claude a lot day to day. I try to use it sensibly, making me more productive and produce better work. I'm also trying not to lose understanding along the way. I want to be able to actually talk to the conclusions I'm reaching.
I have colleagues that seem perfectly content to delegate too much to the agents, and it saddens me. It feels like there will be swaths of engineers that didn't train some of the critical thinking skills that I take for granted.
I certainly see it in slack discourse around anything more complicated than a feature implementation. Maybe I'm just cynical. Time will tell, I suppose.
This is tangential: and forgive my ignorance here, but is there an inherent reason why there aren't smaller, focused models from the frontier model providers?
I'm thinking something like a software-specific subset of Opus that is the default for use in Claude Code. Smaller, cheaper to deploy and consume, maybe faster.
Thanks, I'll have to continue experimenting. I just ran this model Qwen3.6-35B-A3B-GGUF:UD-Q4_K_XL and it works, but if gemini is to be believed this is saturating too much VRAM to use for chat context.
How did you land on that model? Hard to tell if I should be a) going to 3.5, b) going to fewer parameters, c) going to a different quantization/variant.
I didn't consider those other flags either, cool.
Are you having good luck with any particular harnesses or other tooling?
Thank you for all this, I'll give it a shot. Out of curiosity, are there any resources that sort of spell this out already? i.e., not requiring a comment like this to navigate.
> nothing you can run locally, on that machine anyways, is going to compare with Opus
Definitely not expecting that. Just wanted to find a setup that individuals were content with using a coding harness and a model that is usable locally.
What does your setup look like? Model, harness, etc.
Thanks! These things you're mentioning like "You may be able to offload some layers to GPU...", "You can keep the KV cache on GPU..." configured as part of the llama.cpp? I wouldn't know what to prompt with or how to evaluate "correctness" (outside of literally feeding your comment into claude and seeing what happens).
Aside: what is your tooling setup? Which harness you're using (if any), what's running the inference and where, what runs in WSL vs Windows, etc.
I struggle to even ask the right questions about the workflow and environment.
I've been using Claude Code regularly at work for several months, and I successfully used it for a small personal project (a website) not long ago. Last weekend, I explored self-hosting for the first time.
Does anyone have a similar experience of having thoroughly used CC/Codex/whatever and also have an analogous self-hosted setup that they're somewhat happy with? I'm struggling a bit.
I have 32GB of DDR5 (seems inadequate nowadays), an AMD 7800X3D, and an RTX 4090. I'm using Windows but I have WSL enabled.
I tried a few combinations of ollama, docker desktop model runner, pi-coding-agent and opencode; and for models, I think I tried a few variants each of Gemma 4, Qwen, GLM-5.1. My "baseline" RAM usage was so high from the handful of regular applications that IIRC it wasn't enough to use the best models; e.g., I couldn't run Gemma4-31B.
Things work okay in a Windows-only setup, though the agent struggled to get file paths correct. I did have some success running pi/opencode in WSL and running ollama and the model via docker desktop.
In terms of actual performance, it was painfully slow compared to the throughput I'm used to from CC, and the tooling didn't feel as good as the CC harness. Admittedly I didn't spend long enough actually using it after fiddling with setup for so long, it was at least a fun experiment.
OpenNext does model [0] incremental static regeneration, but beyond that I actually don't know, or at least don't recall. OpenNext doesn't do per-route lambdas like vercel does, so it's not like you get any behavior differences there.
I _think_ you can get scale to zero on Lambda by deploying a docker container, too.
1) I don't think it's related to Next, per se, but there may be behavior I didn't build the expertise to comment on. I also know that there were major inefficiencies in the application, so, for example, our P90 latency was (imo) terrible.
2) We'd have to define what constitutes low traffic vs any other arbitrary measure, so it's moot to discuss like this; all I said it wasn't high traffic. You could run it for cheaper, but there wasn't much expertise for self-hosting, for example.
3) For all I remember it may have been half that in daily cost. In any case, miniscule compared to engineer time. What was worse was the prior decision to use serverless aurora rds, that dwarfed everything else in AWS cost - I know this is tangentially related, just saying optimizing that a bit more was not the highest priority, we could do it for cheaper.
It offers packaging for deploying to a serverless environment (e.g. Lambda) analogous to how Vercel does it.
The last question is salient, and it's possible for OpenNext to break and have to catch up to changes in Next.js, though I believe there's some more direct collaboration. I'd say that's the biggest downside - it's not guaranteed compatibility.
I did a migration recently (comments elsewhere in this post), and I don't recall the specific issue, but I _do_ recall running into at least one scenario where OpenNext had made a decision that impacted - in a way that was visible to me and undesirable - how Next.js functioned. That's not a criticism, there's tradeoffs.
OpenNext is just for packaging the Next.js build artifacts. The infrastructure is defined by projects that deploy those artifacts, examples here: https://opennext.js.org/aws/get_started
Some of them are, for example, Terraform projects that list the specific infra. I have experience with the SST deployment, whose website unfortunately doesn't do a great job of listing the infra architecture.
The biggest cost for us on Vercel (several hundred dollars a month) was Image Optimization, and that was because the app was being majorly inefficient with images, in part due to some default behavior in Next.js that we found unfriendly [0], and in part due to negligence. That being said, it wasn't "cheap" by any means outside of that, still hundreds a month for something that I would not consider a high traffic application (I wish I could remember more specific numbers).
Migrating to OpenNext using SST, I think we got the bills for compute and asset serving down to like $15/day or something (granted, we spent expensive engineer time on the migration).
In the company I just left, I actually went through the process two or so months ago of migrating their Vercel deployment to AWS. I evaluated several options that are listed on the website and on GitHub, and we landed on using OpenNext via SST, it was a low-pain effort, especially given the CTO's desire to also migrate off of Next.js.
As other commenters have touched on - my understanding is the purpose of OpenNext is to package the output artifacts of a Next build in a way that can be deployed to a serverless environment, analogous to how Vercel does it. The supporting projects like SST and the other links in the repo are to take those OpenNext artifacts and deploy them to infrastructure generally in an opinionated way - additionally supporting some of the "extra" features described in the repository.
The last project I was working on was to then migrate from SST to Fargate, as a persistent process (serverful?) deployment was preferable for various reasons. In that scenario, we would just be running the built in server using the Next.js standalone deployment mode (effectively a `node index.js`). We didn't need the extra functionality covered by OpenNext.
What does "mainstream" refer to when we're talking about software development and LLMs? As opposed to "engineers".