Has anyone else experienced a significant drop in the quality of their GitHub Copilot completions and suggestions in the last week or two? I rely on it heavily for my daily workflow, and it feels like a switch was flipped. My normally helpful assistant has become… well, a bit of a liability.
I’m noticing a few very specific patterns that are new and frustrating:
* **Excessive verbosity and over-explaining:** It’s suddenly adding long, unnecessary comments for trivial lines of code I would never comment, like `// Set the variable to true` above a simple `let isEnabled = true;`. It’s polluting the codebase.
* **Ignoring clear context:** I’ll be working in a well-defined function, referencing other variables in the file, and it will suggest completions that use completely different naming conventions or introduce out-of-scope variables. It feels less “aware” of the immediate code block.
* **Odd formatting and syntax choices:** I’m getting suggestions with strange indentation or, in Markdown files, it’s suddenly adding atypical YAML frontmatter or formatting that doesn’t match my project’s established style guides.
* **Regression in simple logic:** Most concerning is that for straightforward boilerplate (like a React component prop destructuring or a simple `map` function), it’s now offering more convoluted or even incorrect syntax than it did a month ago.
I’ve done the standard troubleshooting—restarted VS Code, toggled the extension, checked my settings (`editor.inlineSuggest.enabled`, etc.)—but the issue persists. My workflow is now peppered with more corrections and manual typing, which defeats the purpose.
I’m curious if this is a widespread model update issue or something localized. If you’ve noticed similar behavior, could you share:
* Your primary language/framework?
* The most common type of “bad” suggestion you’re getting?
* Any settings or extensions you found that mitigate it?
Hoping we can crowdsource some clarity—or at least some solidarity!
—Hannah
Measure twice, automate once.
I saw chatter about this on another forum. The usual pattern with these black-box AI tools is they push a new model and break existing workflows. You're not imagining it.
There's a bigger issue here, which is relying on an opaque, changing system you can't pin or fork. What happens when it suggests a subtle security flaw during one of these regression phases? You can't roll back to last week's model. You're stuck.
So yes, it's probably worse. But the real problem is building a dependency on something you can't control.
Don't panic, have a rollback plan.