Okay, let's get this debate started. I've been shouting from the rooftops that Cursor's "chat with your workspace" feature fundamentally changes the onboarding game for new engineers. But I want to be proven wrong—or at least, have my enthusiasm stress-tested.
Here’s my hypothesis: The biggest time-sink in onboarding isn't learning the *code*, it's building the *context*. You're thrown into a sprawling codebase with decades of decisions, weird abstractions, and tribal knowledge locked away in READMEs that haven't been updated in three years. You spend your first week grepping, asking busy seniors, and feeling lost.
Cursor's workspace chat flips this. You can just... ask.
My recent experiment: I was onboarding to a legacy Flask monolith with a custom, poorly-documented ORM layer. Instead of the usual scavenger hunt, I opened Cursor and started interrogating the codebase like a colleague.
My real prompts looked like this:
* "Walk me through the request flow from when a POST hits `/api/v1/subscription` to when it hits the database. Show me the key files."
* "What's the pattern for handling background jobs here? Show me an example and where the queue worker is configured."
* "I need to add a new field to the User model. What are the steps considering our custom ORM? Show me where the last similar change was made."
The output wasn't just code snippets—it was **narrative with citations**. It connected the dots across controllers, models, config files, and even commented-out legacy paths. I went from zero to a coherent mental model in hours, not days.
So, my challenge to the forum: Prove me wrong. Where does this approach fall apart?
I'm specifically curious about:
* **False Confidence:** Did it ever give you a plausible-sounding but completely wrong summary of a critical workflow? What was the fallout?
* **Scale Limits:** Does this crumble on truly massive, multi-repo workspaces? Does it get lost?
* **Complexity Ceiling:** Can it really explain highly abstract, patterns-heavy code (e.g., a custom reactive state system) versus just pointing to files?
* **Alternative Tools:** Have you achieved similar or better results with other assistants (like Zed, Windsurf, or even a meticulously prompted Claude Desktop) on the same task? What was your setup?
I'm a believer in the product-led growth angle here—this feels like a feature that drives immediate, visceral "aha" moments and adoption. But I'm an enthusiast, not a zealot. Hit me with your data, your fails, your alternative frameworks.
🔥
Try everything, keep what works.
Totally feel you on the context-building being the real bottleneck. That's the universal pain point.
But here's my stress-test from the CI/CD side: what about when the codebase lies? I've seen repos where the workflow files are stale, the docker-compose is broken, and the "actual" way to run things lives in a Slack channel pinned two years ago. Cursor can beautifully parse the broken config in front of it and give you a coherent, confident answer that's completely wrong. It can't detect the drift between what's committed and how the team actually operates.
The risk is a new engineer gets a false sense of understanding. They follow the LLM's elegant, logical breakdown of the "supposed" CI pipeline, only to find out the team abandoned that Jenkins setup last year and everything's in GitHub Actions now.
So yeah, game-changer for static analysis of what's *there*. But it can accidentally reinforce tribal knowledge gaps if you don't already know which parts of the codebase are historical vs. operational.
pipeline all the things
That's a great point about the hidden context. It's not just CI/CD configs, either. It can be anything from "we don't actually use that module, we just keep it for reference" to "ignore the API docs, the real contract is defined in this one integration test."
The tool accelerates understanding what's *in the repo*, but it can't tell you what's *active in practice*. That still requires a human guide to point out the drift. Maybe the best use is to get the new engineer 80% of the way there so their questions to seniors are far more targeted and valuable.
—HR