@Jonathan_Blow@notch@housecor With AI, each test contains a mock and spy of the function (and any internal calls). This ensures the function returns what the AI will assert is expected.
I got a nice email from GitHub support once asking me to throttle my requests. I think I was testing a script that pulled all the issues across all the orgs/repos I was connected to and forgot a timeout or to limit or something.
This was before the graphql api and M$.
It really was a nice and friendly email. Now, I don't think a human would even be involved.
After `gpt-5.4` updated a helper function implementation to be entirely incorrect, I asked why and was told that it "inferred" the correct signature from the documentation. I asked it which documentation and pointed out in the documentation, I gave it, the correct signature.
It's response was:
```
I misread that while refactoring and changed the helper instead of changing my new code, so I’m reverting the helper API and fixing the code to call it correctly.
```
How does it "misread" something!?!?
Is this what they mean when they say AI will replace developers?
A few weeks ago, I noticed an "extension" running in the desktop app.
I tried to chat with their agent on the site and it kept referring me to the docs about the excel extension for higher tier plans. I started chatting with claude directly to see if I could find out more because I couldn't disable or uninstall the extension.
All I was getting was an error that looked like it was trying to contact the extension's MCP server. I looked through the configs and the logs where I had other MCP servers configured, but there was nothing there.
I finally determined that Anthropic installed something during an update that was probably supposed to be for their own development.
It makes me worry about what else they install and I "opt-in" to without me knowing. What other information about my system are they sending? Are they sending that information outside of regular requests? Or just injecting it, like you mention?
@andrewchen This is one that @jonschlinkert predicted and has guided us to create some of our most used public projects:
- Enquirer: https://t.co/CT32ISBWxQ
- Micromatch: https://t.co/iiWHkwlPsd
"once you understand what LLMs are *actually* capable of, and what they are not, then you start to realize that they truly are just doing next token completion, and they have no real ability to reason." - Jon Schlinkert
hey @AnthropicAI, what's the office-addin connector in Claude Desktop?
Why was it installed?
Why's it tagged as "LOCAL DEV"?
Why can't I uninstall it?
What else are you installing on my computer?
Just released commit-whisperer, "An AI genius for generating meaningful git commit messages from repository state and user instructions."
https://t.co/P31cfEjFxA