I used it on a data science project to find the best rules for achieving a defined outcome. At first, for fun, then I actually used some of its insights (and it caught a sampling issue I overlooked, oops)

I was surprised: even tho it was the cheapest option (against other small models from Anthropic) it performed the best in my benchmarks.

I'm a designer who works with Claude Code daily. The standard feedback loop is frustrating: I'd screenshot something, describe it in words, and the AI would misinterpret what I meant.

So I built a Chrome extension that lets you pin comments directly on live page elements, then export everything as a structured prompt. Click an element, describe what you want changed, hit export, paste into your AI agent. There's also an ASCII canvas mode for roughing out layouts and edit existing elements. No images, everything stays as text that fits in a clipboard. Local-only, no accounts, no cloud.

This is one of my first "dev" projects, would love to hear what you think!

Comments URL: https://news.ycombinator.com/item?id=47235860

Points: 2

# Comments: 0

As a designer I'm having a lot of success vibe coding small use cases, like an alternative to lovable to prototype in my design system and share prototypes easily.

All the devs I work with use cursor, one of them (front) told me most of the code is written by AI. In the real world agentic coding is used massively

Lately, I started using Perplexity Discover a bit more, and it got me curious about similar solutions to curate and summarize news based on what I find important, not what everyone is trying to push down my throat.

How do you keep yourself informed? AI curation? Trusted publications? Social media follows?

I feel we have all the bricks to have a great experience of learning about what's happening in the world, but instead we're drowning in clickbait articles, propaganda pieces, and even previously reputable publications have the quality of their work taking a steep dive.

Comments URL: https://news.ycombinator.com/item?id=46345011

Points: 2

# Comments: 6

Even if it's not benchmark, a vibe test from a trusted professionnal with a close use case to mine would suffice.

Your point about ecosystem is true, I just switched main main provider from OpenAI to Anthropic because they continue to prove they have a good concrete vision about AI

- Augmenting CLI with specific knowledge and processes: I love the ability to work on my files, but I can only call a smart generalist to do the work. With skills if I want, say, a design review, I can write the process, what I'm looking for, and design principles I want to highlight rather than the average of every blog post about UX. I created custom gems/projects before (with PDFs of all my notes), but I couldn't replicate that on CLIs.

- Great way to build your library of prompts and build on it: In my org everyone is experimenting with AI but it's hard to document and share good processes and tools. With this, the copywriters can work on a "tone of voice" skill, the UX writers can extend it with an "Interface microcopy" skill, and I can add both to my "design review" agent.

I'm still wondering about how this approach would work when conceptualizing flows, systems, or more complex interfaces

I could see a sponsored section in the middle of the reply where the LLM just tells of these vendors align with what the user is looking for

If you're tucked in right behind the absolute frontier models, the economics change completely

I use OpenCode to try out various models, but haven't really landed on one specific one. These days, I use Gemini even tho I never know if it'll get stuck or not.

I'm curious if others have experience with CLI / agentic workflows for non-coding work. I'd love to hear about your experience with the various models and how you use them.

So far, my + and - are:

    +: Finally, control over the AI output, no more copy-pasting everything, contextualize every new chat. It's also way better to organize a task for the AI in Obsydian than in a chat interface.

    +: Agentic workflow is interesting, I can throw a task, go back to my work and review later

    +: API means you can keep your costs low if you're not a power user, it also means you can easily switch providers for different tasks

    +: I feel like a hacker

    -: It gets pretty complex when you want to expand the AI capabilities. Setting up RAG, MCP looks a bit daunting. You can create "agents," but they are basically a system prompt. OpenAI custom GPTs are so much easier to set up. Maybe, given the current size of the context window, I'm overthinking it.

    -: Initial setup was also not the smoothest, but I'm rusty in this area and Windows is not helping

Comments URL: https://news.ycombinator.com/item?id=46061841

Points: 2

# Comments: 1

F is not set as an alias by default tho

Small point for Flow here again, because you just have to use the prefix doc: to search through your files, whereas on Raycast, you need to set up an alias and enter the extension. Both have file preview

- Raycast has a nice UI that can expand to work well with extensions

- Flow is faster to use. With Raycast you often need to enter an extension to finish your action. To launch a scrip on Flow I just type "r [shortcut] -> enter" while Raycast is "quicklinks -> enter -> [shortcut] -> enter. [edit, with minimal setup using aliases, you can have similar speed. See __jonas comment below]

- Performance-wise, Raycast was often eating my RAM, but a dev mentioned it's expected in the beta, they'll fix it for the launch. Otherwise, both feel snappy

- Both seem to have enough community support and extensions

- I never really tried the AI features, I don't know if it's the right place for me to augment my workflow w/ it

Curious about the experience of others with these tools or similar ones