- Autoplay transitioned to a per-device setting… suddenly default-on on every new device you log in to.

- Watching a video with computer-to-TV account connection? Automatic “TV queue,” a concept absent from the TV app, with incomprehensible behavior for how it’s used, so now videos are auto-played anyway.

- Watching a video from a playlist? Autoplay cannot be turned off.

Is it simply bad/absent product management and product design? Or is it actively user-hostile decisions meant to prop up view numbers and continue to have the users hooked on YouTube?

I’ve tried so many things and I have to beat Opus 5/Fable 5 over their head every time. With Opus 5, the failure to actually improve the writing is borderline comical. Both models have been building up tomes of memories on top of my core rules, all to be ignored.

Nothing sticks mid-writing! The only lever is to ask to revise after the fact, a particularly futile proposition for anything non-trivial with Opus 5.

In contrast, GPT 5.6 Sol Max absolutely obeys my edicts to write well. I selected the concise writing style and its default writing is really not bad, but tighten it up with a standing AGENTS.md order and it obeys.

The downside is that even at Max, it’s not as good as Fable or Opus 5 xhigh at writing code. Larger work and the 256k context’s forced compactions cause it to lose track/fidelity of critical details. Review passes are essential.

Another reason I’m considering leaving Anthropic are the sporadic refusals… Fable printed a Markdown body with a hex dump to inspect for trailing white space and line endings, boom, denied due to `reasoning_extraction`. Had to tell Fable to never do hex dumps. Asked it a few times “what do you think is typically done for this?”, got another `reasoning_extraction` error. It forces you to lose a whole turn of work when you have to press Esc/Esc to “retry” the last prompt… except if that prompt was mid-turn, you’re losing your entire turn.

The recent BashFirst experiment is hella nuts, where it prefers writing bash over Edit/Write tools in auto mode. WTF Anthropic?

I pity those stuck with Claude in an enterprise/work setting.

Also it looks like English might not be the author’s first language—might not lead to the best training corpus.

Don’t waste your time rewriting it by hand just because keyboard warriors are up in arms on HN.

The material feedback here reflects the questions who might be from someone who’s been running local LLMs, so ask the LLM to update it with that in mind.

Another poster pointed out it’s not clear what the quantization is. There’s that one paragraph but it’s confusing. What I want to know right away is: are you running the unquantized model or is it quantized? If so, how much? Use terms like Q3 or Q4 or 4-bit or 8-bit. Explain why it’s not practical to quantize less. How much precision loss do you think there is at the quantization selected? What if I had 128GB RAM and wanted to have better precision and not higher token speed—would it be a good idea to choose 4-bit instead of 3-bit for some of these layers?

Funnily enough, most anti-slop skills I found are both way too verbose and miss some common slop constructs.

I also reduced many rules from “When doing X, don’t do Y, but do Z.” Instead, the rule is “When doing X, do Z.” Fewer tokens and often works better.

I had one critical rule I was maintaining about searching the codebase using a structural index/graph and not grep. Every time the agent missed it, I asked it how to improve the rules. Eventually, I asked the AI to review that rule file and it rewrote it to be 30% smaller, but, crucially, structured to be more understandable by the LLM.

Another helpful thing was to ask AI to review my rules for things it can load on-demand when it works in that area.

Example:

https://snyk.io/blog/node-gyp-supply-chain-compromise-self-p...

Before that we had event-stream, then we had XZ compromise.

It’s not exceptionally hard to delay reaching out to external sites until after a cooldown period.

The practice of storing secrets in a .gitignore'd .env.local.json or whatever is a really bad idea and I can't believe that it has become a normalized, acceptable practice in the industry.

Ultimately this combo worked:

1. https://pi.dev/packages/pi-tool-guard —- corrects key name synonyms and common structure errors, so tool calls succeed automatically (e.g if the model hallucinates old_str instead of oldText). It also wraps top level oldText/newText in an edits array if the tool didn’t do it.

2. https://pi.dev/packages/@aboutlo/pi-smart-edit - white-space-tolerant edits, as Qwen would sometimes add a fifth space to a four space indent

Hashline edit tools didn’t work well for me at all, they confused the model and it still failed to edit correctly. Also line removals would invalidate the rest of the file requiring re-reads. I tried pi-hashline-edit-pro, though I see it now keeps a database of hashes to help keep them stable across edits. Regardless Qwen kept thinking that the hashline prefixes were part of the source.

When you are talking about checking your dependencies in the source tree, you are effectively pinning exact versions, and not using floating/tilde versioning syntax.

Turns out there is no equivalent to “npm ci” that doesn’t clear node_modules first, and you can’t call npm install to simulate NPM ci behavior (sans clean).