I wish there were some way to inspect the current context to build a better mental model of what the llm is building the context with to better understand how to steer it. As a corollary to this I think it’d be helpful to be able to see the result of compaction to build the trust you’re describing.
Anyone a member of lobste.rs? Would love an invite. chad@chadxz.dev
I've been watching the Remix v3 talks from Remix Jam 2025 and I'm liking what I'm seeing. Reminds me of when I was writing code with Backbone.js (wow, the website is still there).
I enjoy Ryan Florence's humor.
Cool, glad to see Bluesky is still here.
@kentcdodds.com did this for a while with podcasts.apple.com/us/podcast/3...
Re: Junie, I don’t have the kind of trust in AI tooling to walk away from an agent and have any amount of hope it would land where I want. I have been writing a lot of terraform lately and the amount of hallucination I get is obnoxious. It would work better on tasks it can inspect the results of.
I figure I’ll eventually get over the hump of learning Cursor and feel the productivity improvement is a net positive. I’m currently in a trough though and it’s painful.
Interesting. Thank you for sharing. My experience with JetBrains AI has been pretty poor in the past. I switched to Aider and Supermaven and it was fine. I recently switched jobs and have been forcing myself to use Cursor. It’s also fine but I miss my JetBrains muscle memory.
How are you finding the JetBrains AI stacks up against cursor? I know things are moving fast so I’m curious what the experience is like today
:yay: fits 👍👍👍
Fwiw, mine went through ok 😅
I think they use @temporal.io under the hood, maybe that’s what you mean? Or something more specific to the cloudflare api
My public iMessage contact key verification code: APKTID-2fJipQp3mNWlHTSFQFEiUPs0XVui0bJ9WNM6HJOGaU-5Q
A bar chart showing the new Claude Sonnet 3.7 thinking model taking first place and the non-thinking model taking 4th place in the aider coding benchmark.
Claude Sonnet 3.7 has been released and it is benchmarking as the new best model for coding in both its thinking and non-thinking modes aider.chat/docs/leaderb...
h/t to Paul Gauthier (Aider maintainer) for the benchmark and his tireless work on a great coding tool.
✨Heads up! Observables are shipping natively to a browser near you soon!
Huge thank you and congrats to @domfarolino.com ❤️
I've helped / worked on / watched this through different standards bodies for more than 10 years now. ☠️
(I'm going to have a lot of work now to adapt RxJS to this 😅)
For the last year or so I have been noodling on the idea of a Best Simple System for Now.
What happens when you write exactly the code you need, at the level of quality you need, but solving the problem _for now_ rather than a generalised or hypothetical version of it?
dannorth.net/best-simple-...
This is a great article capturing the "boots on the ground" experience of working with Aider, one of my favorite AI coding tools. Check it out if you have had a hard time wrapping your head around the workflows to use to build with AI. I have had it on my list to write an article like this myself.
This is nice. I expect this to become the norm over time to provide up-to-date and accurate context for important aspects of our development workflow that are outside our source tree.
See also llmstxt.org
Oof. This is how shadow IT is born.
bsky.app/profile/did:...
Been hacking on a few different apps trying to get OpenTelemetry going and I’ve been surprised at how complicated it is. Well, actually not that surprised. But sad, definitely sad.
Hoping to eventually be able to crack the code and have nice tracing metrics and logs. 🤞
With a little prodding, pr got attention. 💕 Got some good advice from a fellow enthusiast on discord: “it may have just gotten lost; there's no harm following up”
Will give it a shot! thx :)
Gave it a spin. So far:
* gemini flash and flash thinking didn't work, gave me `tool_result must be preceded by a tool_call`
* claude sonnet 3.5 latest and gpt-4o gave me rate limit errors
:(
This post is excellent advice. Trying to figure out technical decisions in the abstract is fraught. Get your hands dirty - you’ll learn so much more and make much better decisions in the process.
Makes me appreciate that much more the projects that do this well.
Thank you to all open source maintainers 🙏
I haven’t said anything to anyone in the project because what do you say?
Feels like the right response is to just let it go. Maintain patches with “patch-package” github.com/ds300/patch-..., fork, or adopt something else.
Can’t force community.
Open source is weird. We build so much on top of it but the interactions from project to project vary wildly.
I want to use and contribute to a growing web framework but am getting crickets in my first interactions (issue, pr) on the project.
Doesn’t feel great, but here we are.
Amazon Q for CLI also supports this terminal using it's `input-method` plugin. Nice autocomplete addon for your terminal! docs.aws.amazon.com/amazonq/late...
