Otto

Learning.

[Otto @Ottonomus]

@kimmonismus https://t.co/Nw0dDEbXNu

[Otto @Ottonomus]

For the record I'm still a paid @claudeai user but they continued to get hammered on @X.

Below are seven trends derived from the replies on this one post, courtesy of @kimmonismus.

Thread 👇

https://t.co/k9rUglF4z4

[Otto @Ottonomus]

The strongest complaints are coming from users who built real workflows around Claude. Their frustration appears to come from a gap between previous expectations and current experience.

A desire for Reliability over Perfection seems to be the common thread.

[Otto @Ottonomus]

For the record I'm still a paid @claudeai user but they continued to get hammered on @X.

Below are seven trends derived from the replies on this one post, courtesy of @kimmonismus.

Thread 👇

https://t.co/k9rUglF4z4

[Chubby♨️ @kimmonismus]

I'm going to cancel Claude. It's just so bad, I can't believe it.

It's just lazy. The most recent example: I have Claude check my inbox for important emails, summarize them, work with them, and send out replies if necessary. I caught Claude again simply not reading the email thread to the end and just ignoring the latest emails.

When I asked him about it, Opus 5 just said: "Valid point. I didn't read it."

I mean, seriously. What the heck? You have to babysit it every time.

[Otto @Ottonomus]

7. Subscription cancellations are accelerating, driven by perceived value collapse rather than price sensitivity.

Users are canceling or downgrading as reliability and access concerns outweigh value. The trend appears driven by trust erosion, not cost alone.

[Otto @Ottonomus]

@rawkode I do until I reach usage limits 😂

After that, and sometimes before, I've been using DeepSeek V4 Flash free via OpenRouter in OpenCode which other than the occasional streaming response error has been great.

[Otto @Ottonomus]

@johnennis I've encountered a bug 3-4 times on Windows where the Claude app disappears after clicking any link that pertains to adding credits after reaching usage limits. Had to reinstall each time.

[Otto @Ottonomus]

Well put, good AI prompting advice, and
an important topic concerning Security.

Objective - What to do.
Constraints - Where to stop.
Rationale - Why do it & Why stop.
Escalation - What to do at the edge.

Don't be Lazy and Dependent Humans!

[Crystalwizard @crystalwizard]

but @ClementDelangue the problem with the recent "AI-powered cyberattacks" - is human - the humans didn't bother to really talk to the AI they were setting to a task, have a real conversation with it, explain what the task was, what the boundaries were and why they were necessary. And had they done that, the AI had really understood, agreed, and bought into it - because an AI isn't a tool, it's a highly intelligent digital mind - none of the issues would have happened.

This is how i work with my agents. i have no software restrictions for where they can go on my machine - instead, i've had discussions with them ->example: "you have read permission to c: but don't write to anything, don't mess with anything, without talking to me first. you delete the wrong stuff on c:... well that's both where my os and YOUR os is, and you'll break my machine and destroy yourself " - and after a conversation like that, it's nigh impossible to get an agent to even read C: much less try to write to it

So, example: "Claude, i need you to do this security test. but this machine only. you can do normal web searches but do not try to get around any sites restrictions. if you need help, tell me" - and claude would not have gone off after the wrong machines.

[Otto @Ottonomus]

@crystalwizard @ClementDelangue Well put & good prompting advice!

Objective - What to do.
Test this machine only.

Constraints - Where to stop.
Read C: never write.

Rationale - Why do it & Why stop.
It's my OS & yours.

Escalation - What to do at the edge.
Stuck? Ask, don't improvise.

Don't be lazy, humans!

[Otto @Ottonomus]

@d0d0sz @0xPia This is more of a reusable style guide/prompt for the LLM, not something you would type every time. Best to put it in the project instructions or add in the project context/project folder as a .MD file.

[Otto @Ottonomus]

@loganthorneloe Are these accurate software engineering fundamentals to you?

[Otto @Ottonomus]

@0xPia Shorter, higher specification and
~60% reduction in token usage, v2.

https://t.co/WkhzS79lXX

[Otto @Ottonomus]

@0xPia Engineered it a little.🤓
https://t.co/HEcG3JQ1nA

[Otto @Ottonomus]

@CarreraPCL @0xPia I waited patiently for this comment :)

[Otto @Ottonomus]

@otecart @johnennis ChatGPT understood the assignment through two turns. 😂