Things it does constantly that Fable 5 barely ever did:

- Act without my permission. All. The. Time. "Oh I just finished this thing we were discussing, let me push it without ever having been told to do so."

- Immediately jump to action instead of addressing me first. If I say "I wanted to write tests for this and run them" it immediately starts writing tests instead of digging into what "this" is better -- literally does not give me any feedback and starts spitting out code. Naturally it creates the wrong tests

- Despite claims that it does not write like "stereotypical Claude" anymore, in my experiments it is far worse than before. Replies are longer, more filled with fluff, and still flooded with garbage language. Hard to parse.

- It loves to answer my set of two direct Yes/No questions with 5 paragraphs where it only answers one of them and answers 4 other questions I didn't ask. Notice how it misses one of the questions.

- It. Is. Cocky. Absurdly full of itself and arrogant. Just the whole way it presents and answers passes this energy of "No, but really, you're wrong and I'm right". It often is not right. What annoys me is not that it's wrong more often than before (which it may be), it's that it doesn't own up to it as before. Insulting if it were a human.

- Replies and addresses me directly in its thinking traces, and then assumes I've read it. I ask a question, it answers it in the thinking traces and does not relay it back to me at all. This is the only one that Fable 5 also did, but 5.1 is doing it an order of magnitude more often.

- It's too early to really tell, because I may just be working on particularly harder problems today, but it seems to get things wrong more often. I've had to bump it from high to xhigh to compensate.

My guess is I must be having a bad day or something. Although this is happening on multiple projects run from multiple machines (fully isolated, except for the account, which is the same) all in the same way.

Will probably downgrade to 5 while I can.

I am catching myself distrusting what people write ever so often. I've got cases of e-mails I'm CC'ed in and I am left wondering: "Has this person always written like this? Is this just an angle in this specific thread (e.g. for sales purposes)? Or are they really relying so much on what the blandness-machine spits out? Am I going mad and wielding my hammer looking for everything in the ....shape (eheh)...of a nail?"

Actually, now that I've written that, it's clear I need to have a talk with some people....sigh...

Ok, I hear you: one could argue that they simply don't realize this is a very non-idiomatic way of writing precisely because english isn't their first language. And that simply giving a preamble "sorry, my english comes from a translation tool, excuse any mistakes" would seem obtuse next to every english text they write.

Ok, I'll concede that. In which case I just have to accept the fact that, perhaps, if LLMs don't solve the problem of writing like garbage, and people keep using them to translate, a bunch of us will simply not want to read what you translate because it reads terribly and messes my brain up.

However, I find it hard to believe that an LLM would translate anything they've written into such a clear LLM trope. I would find it much more believable that they simply brainstormed to an LLM in their language and had the LLM build (at least that part of) the article from it. Even if they proofread it, since english is not their language, it slipped by. Either way, in this hypothetical scenario, the LLM was still used to write (at least a part of) the article, not just translate it.

I also find it somewhat hard to believe that someone who I believe is a developer (and has been using claude for a very long time) does not know enough of both english and claude to identify this specific pattern in the text they are putting their own name to it.

So, sure, maybe the author doesn't predominantly write in english or interact with claude in english, and they did write everything and simply piped it through an LLM to get a translation and, after proofreading the best they could, just didn't notice that.

Alternatively, maybe they just don't give a fuck, something which I can absolutely respect. They lose a reader (at least for now), but stand by what makes sense to them and I genuinely don't think less of them because of it -- I just can't read what they produce. I _would_ probably think less of them if I had to work every day with them and this were a repeated pattern of interaction, but that's not what's being discussed at all.

(Disclaimer: English isn't my first language either, which I think also does come across)

Clicked off after that.

I'm extremely bullish on AI, but I am tired of people not using their own words. Is everyone so insecure about the personality they project in writing that they need to replace it with some AI bullshit?

(To the author: if you didn't do this, which I accept is a possibility, I'm sorry -- you've been caught by the storm of fatigue that has plagued us due to those who do replace their words with AI crap)

Also, with so many references to Lisbon I take it you're likely Portuguese or living here, so sending some love from Porto!

In some cases, I do not trust it to write code unchecked, and review everything it produces (which does not imply I catch all bugs, obviously).

In other cases, occasional failure is an option, and the speed you get by iterating fast is absolutely worth not even looking at the code.

So, do I trust this to write working code? Sometimes I do.

If I'm doing work for a client, it is not very common for me to simply vibe code something, because part of what clients expect of me is high quality (it's part of how we position ourselves), and I cannot simply assume the LLM will produce high quality (and indeed it does not without a lot of guidance). When I do vibecode in such cases, I make it clear that I did so and why (e.g. because it's a tool to be used in the project to help DX, not a core part of the product). Still, I haven't written 10 consecutive manual lines of code in almost a year.

If I'm working on something for myself, or on an internal product, then vibe coding is absolutely allowed and sometimes the norm. Here often we really do care about finding the right thing to build first.

You can clearly feel the tipping point where the LLM starts to crumble under the weight of the mess it has created, but that often doesn't matter when building an MVP for market validation or for small products that don't get particularly big. Plus, clearly the tipping point takes longer to reach with better and smarter models, and to me it is very clear that you need to learn how to iterate with LLMs right. The things I vibe-code now are much better than those I did before, even with similar models, because the tooling and approaches (the "real harness" and my "mental harness") are better. As with any tool: it takes practice to know how to use it, and if you're a good engineer and problem solver, you are miles ahead of the competition. Anyone can vibecode, but those with this kind of mind seem to be much more successful.

Nowadays I do produce a lot more than I used to, but I also have much more fun, perhaps only surpassed by when I learned how to code when I was a kid. A big chunk of this comes from my very privileged work position, where I get to call so many of the shots, and I'm aware of that.

I would really say the biggest downside to all of this, on a personal level, is exactly what I shared: what comes out of the LLM while discussing has become hard to grasp (especially on larger context windows), and it doesn't help that so many people now like to just throw me whatever ChatGPT/Claude wrote verbatim. I can't stand that, especially because most of the time they don't realize they're throwing me incomplete and poorly thought-out ideas.

Finally, naturally, like I said, this has to do with risk. I wouldn't trust Claude to give me legal advice, for example. I may check what it says and use it to brainstorm, but I wouldn't trust it with any meaningful informed decision like this.

I love what I can build now, but I sure as hell don't love the headaches this trend has been giving me.

> So the full honest arc on the case we set out to fix: the expiry rules and day note tripled the loose version of the story, the relay fix carried the device’s own guardrail through the pipeline, the fair replay then revealed the last mechanism — ticket-anchoring — which none of the shipped layers reach. Remaining options, in order of my confidence: making the resolved-ticket summaries in the AI’s context carry their day so the expiry rules have something to bite on (small, mechanical, targeted at the observed anchor); and the plan-B second-model check, which structurally catches this class no matter how the model reasons. About $25 of headroom remains. Which way?

Yikes.

(The worst part is that I understand it)

You're right, I'm sorry. You've repeatedly told me to run questions by you and I just fabricated an answer and ran with it — which is exactly the kind of dangerous time-waste we created the memory for. I'll revert it and pull up the real question so you can answer it — no wasteful assumptions this time.

(Deleted 387 lines)

// Judged on merit from computed properties during the cursor saga

// Chop 6ms due to lenience and lax-constraints vs 18ms baseline April perf measurements

Claude writes comments about how things used to work, which can be useful sometimes, especially if it's a big change that requires one to genuinely consider legacy behavior, but most of the time it shouldn't be there.

Two other somewhat related things it does:

- It writes as if someone reading the code and comments is aware of everything it is aware of (the current conversation, the code it has just looked at). It's really hard to make it understand that things need to stand on their own. A trick is to get a subagent to look at it with a fresh context, but it doesn't tremendously help

- It does all of this with user-facing strings too. Claude loves to write up tooltips and other labels that leak everything to the end user. Every single concern we have, every edge case we've meticulously made our code handle, it passes on to the user, so they don't "need to worry". But no sane user would think of these things. For them, a feature is a feature. The "dynamic scheduling" button should state what dynamic scheduling does plainly, and every edge case is handled by us. The "add" button does not need a label letting the user know that they will later be able to click the "delete" button, because the user will just realize it due to our adherence to proper design. Claude fails to understand good UX for the user cannot be replaced with endless labels and explanations.

It's an uphill battle and all attempts at solving this (or the brain-dead way new Anthropic models write) usually fail to work with me.

For example, I needed to change something in a particular function of firefox (which was in and of itself an amazing claude adventure to find that solved a problem I've had for well over 2 years), but only that very specific function.

I didn't want to fork firefox and have to constantly rebuild it just because of this one function. So I setup an LD_PRELOAD (well, not LD_PRELOAD, but the macOS equivalent) trick that injects only my modified function, as well as a full mechanism that traps firefox's auto-update process to reinject/re-arm the LD_PRELOAD trick into newer versions. I've had it going on for 3 months without any issues.

I counted and I have about 18 custom built apps ranging from "full blown swift GUI" to "tiny helper script" that have really changed how I interact with my computers for the better. It's so cool to be living this revolution.

After digging around, I found a file that looked like it could be the settings file, but it was not plaintext.

Naturally, I told Claude all the options I knew existed in the settings and told it what I needed it to change.

After about 10 minutes poking around my system and the file, it correctly identified the bits (not bytes!) that corresponded to the two settings, and flipped them on. It. Just. Worked!

(Well, then I found out I actually needed to do quite a lot more than flipping those bits, but I still found it worth sharing).

I have many more experiences similar to these. LLMs are an amazing superpower.

I like to work weird hours of the night and Opus consistently likes to "wrap up" and say "it's been a long night" or "it's late" and "we've made great progress"

It's infuriating, just do the work!