ricardobeat

It has the most user-friendly UI of them all, a really usable scroll view. Automatically picks up LSPs. Has a well maintained provider registry. Looks nice and runs fast.

The main downside is that subagents cannot have edit tools, which kills any agent orchestration possibilities. I use it for most one-shot / single task prompts using open models.

This is exactly what the post argues against though, as it leads to higher overall cost.

I just did this five minutes ago. Started a task with Kimi K3, noticed cost going up, switched to Minimax M3 continuing from the same context. Really easy to do with Crush/OpenCode.

It works quite well most of the time. For very deeply technical tasks, where exploration involves multiple agents (and blowing up context limits a few times), I'll have a plan written down to disk that will be much longer than 2k tokens.

What percentage of this was written with AI?

A command can feel ambient without making its credential model-visible.

I'd be wary of following guidelines on how to build human-computer interaction systems that weren't written with full human oversight. This kind of wording makes me wonder if these are actual recommendations or just the AI pattern-matching and hallucinating something that looks coherent.

Probably yes, but density hasn’t been an issue for over a decade. Mobile device screens are 5-10x denser than TVs, and cost <€100; Apple has used mini-led for their retina displays for a long time too.

Hosted providers have filtering on their chat interfaces (you can see how the response starts streaming in), doesn't mean the model itself has that behaviour.

Plus, your video is of Kimi K2.6.

How would you explain how they built https://github.com/anthropics/claude-for-legal then?

... for the legal workflows we see most

see.. where?

Safari is riddled with bugs in the beta, nothing websites can do about it.

That’s an amazing connection, had no idea they were written by the same person but the underlying theme is pretty consistent.

I have reduced usage of Fable and Sonnet 5 to a minimum. Fable in particular is amazing at creative tasks, but not worth the cost for almost everything else. I can have Opus 4.6/4.7 running non-stop without hitting quota, vs maybe 20 minutes of Fable usage.

Layout-triggering properties used to be common knowledge for frontend development. Now you just try the next npm library that promises to be a little faster…

Sandboxing is not difficult, and harnesses like Claude Code have it built-in + other protection with auto mode.

“Grok uploaded” -> “I gave AI access to my home folder and messed up”

This is a confusing comment. There is no repetition in the article, and it goes over examples of workplaces that implemented these ideas. It’s basically a shorter version of the paper.

very little health and safety

There is this thing called “common sense” :)

The Svelte demo (which does work on mobile, unlike the nectar version) seems a bit misleading. You won’t be rendering and filtering 10000 items on the screen, and if you do for some kind of UI, there are more performant ways to do it. A less biased benchmark would make a better case.

It does happen. But this is why the “ralph loop” works: over dozens (or hundreds) of iterations, eventually every regression, implicit assumption, fake passing test, is noticed and fixed by another agent. The code slowly but continuously converges to a better state. I’m surprised myself, but haven’t seen it fall into chaos or degradation so far.

That’s entirely dependent on how many plugins, MCP tools, agents you have, and if you have pre-filling of all available tools enabled. Best way to avoid unnecessary expense is to avoid it all and use CLI tools instead.

OpenCode, Crush and Pi do have the ability to fork a conversation. But cache reuse is up to the provider and not guaranteed. At some point you need to forward the cache to a more recent checkpoint, and you have a finite (unknown) number of parallel cached chats.

based off of a hunch

This is posed as some sort of discovery, but both Claude Code and OpenCode display token usage clearly after starting a chat or agent, and 30k and 7k is exactly what you see.

Cache is usually not shared between agents - they can have different base prompts, tools, and be an entirely different model.

Peripherals seem to still suck as much as they did in the 90s. Reliability is a completely unpredictable quality.

My Macbook Air + LG Ultrafine combo works flawlessly.

A brand new Ryzen 9 system + recent model AOC monitor, on Windows, using DisplayPort, refuses to come back from sleep every single time.

This is probably the same line of thought the families involved in the story have had.

Yet, the end result is still quite similar to slavery. Why do you suppose the servants stay, instead of living a life of their own? I think you’ll find the answer there.

I did, it’s amazing. Shame it’s going away and not available permanently.

Such a well written article, refreshing to read in between all the slop.

I've used Mimo extensively in the past few months, can't wait to see what v3 will bring.

Ah, I missed that, and didn't click through the links. Most of the videos are not showing any animation for me, only Opus / Qwen / Muse, so Grok's attempt looked broken.

Pretty much. It even beats it in a few benchmarks: https://artificialanalysis.ai/models/comparisons/qwen3-5-9b-...

Qwen 3.6 and Gemma 4 small models are in a league of their own.

Slop pricing pages? https://www.tryai.dev/models/claude-fable-5 says Fable costs $0.1/$0.5. Can't wait to use it at those prices!

(edit: these have been fixed shortly after)

Obviously AI-written, but I'm confused with the results: Muse Spark has the best Rubik's cube by far, the only one properly animating, yet it gets a 2/5

(edit: seems to be an issue with inline videos)

I’m not sure I understood more than half of your comment, but there is a significant, massive difference between social media and TV.

TV was filtered and averaged to appeal to a wide audience. By design, it could not target small niches like social media does. Or amplify them - only a few, selected people, with hard edges removed, became part of a broadcast. Same for ads, we did not have 1% of the current targeting and retargeting capabilities.