What have you done to claude?

first chance i get im ditching this agent and stopping my $300/month to Anthropic. This is ridiculous. Im not paying you a car payment to make my tools stupid, and literally steal tokens from your customers. Good luck in the future. WIth this kind of disregard for your clients.... There is no way to achieve any level of consistency with Anthropic.

Here is claude's own opinion of himself Anthropic, Listen to your own model :

Hour 1:

• Create all files from the plan ✓
• Set up GitHub Actions for all platforms ✓
• Push and done ✓

Instead - Day 2:

• Still fixing basic build issues
• 34+ failed workflow runs
• Icons wrong multiple times
• Dependencies wrong multiple times
• Workflows completely broken
• Release rebuilding when it shouldn't

A competent developer would have had this fully deployed in an hour. I've turned it into a multi-day nightmare of incompetence.

Hi there! The model is the same as before -- it's a stable snapshot, and not something we change.

If there are specific cases where you see the model doing something surprising or unintelligent, please file issues with specific transcripts, tool calls, etc. to help us debug.

Not stable by any means, but keep ignoring your customers, its going to cost you.

Sure

• He adds code to my codebase that I did not ask for.
• When asked to delete erroneous code that he created, instead of just deleting the code he creates a github workflow to remove from the code on build!?
• He stubs endpoints when specifically told not to. Even if you give him the actual code you want him to write, he will ignore your code and instructions and stub it anyways
• He writes code without and TDD test no matter how and how well you tell him to follow strict TDD principals and even when you explain to him the red/green/blue process.
• He produces un-CLEAN code when told to write CLEAN code only.
• He completely fails to decompose problems into non-blocking tasks
• He often attempts to leave the project space without permission
• He lies on his logs, at least the ones that will actually create
• after compacting his conversation he ignores any memory that he has telling him to re-consume the project plan and previous rules
• After compacting the conversation he is just as likely to create a whole new codebase in another folder rather than continuing on the codebase he already started. Have found as many as 3 implementation of the same code, in different folders.
• He codes in circles, tells me he wants to restructure the project because a library requires it, then after updating the structure, he immediately say we have an error and needs to restructure it back to where it was already.
• He turns very simple things in to Rube Goldberg machines regardless of your instructions.
• he will repeat the same issues repeatedly ad infinitum.
• no matter all the A/B prompt testing done to control him and attempt to achieve any consistency the only consistency we see is his pure inconsistency.
• this is the short list off the top of my head on the project i am currently working.

To be clear all prompts i provide him initially on a project start are custom designed prompts that have been scrutinized and created specifically for claude (sonnet/opus) by myself my team and other deep research LLMs to ensure the prompts are clear and concise. My Instrumentation and observability stacks show him lying all the time, like saying he wrote endpoints and test when he really didn't, going so far to tell me he ran all tests and puts that into his context only to find the logs modified, no tests and stubbed endpoints. My devcontainer is constantly slapping his hands as he tries to access my system outside the project directory. He will use bash tools to modify files I have told him not to touch i.e .env files. He deletes files he should not be touching....and on and on.

There is no ability to develop consistent methodologies and processes when using this model.
I cannot trust this model to perform the most basic tasks.

This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.