I barely scraped by.
As far as I can tell, there's currently only one question in the question bank — a library book search/borrowing system. There's already a very detailed writeup of this question in another post, so I'll just add a few of my own thoughts here:
For round 1 and round 4, when talking to the AI PM, try to put all your questions into a single prompt and have the AI PM answer everything at once — it saves a lot of time. Before the conversation, quickly read the question description, since there are some hints in there — like the fact that you need to understand user requirements, implementation details, front-end/back-end logic, test strategy, etc. If the AI PM is satisfied with your clarifications, it will tell you at the end that it got everything it needed, and that's when you can end the conversation. There's also a dedicated place to take notes — if you want to save time you can just copy the AI PM's answers straight into the notes.
Round 2 is writing code with Claude. Note that the instructions say you're not allowed to just copy the AI PM's answers directly into Claude. To be safe, I typed out a whole long prompt by hand, which was extremely, extremely time-consuming. I didn't run into any code signal or Claude crashing issues, but running the tests plus having the AI write code was pretty slow.
Round 3 is fixing a bug. You're not allowed to use Claude here, but the bug is very simple — same as what the post above said, you just need to move a piece of code to a different location. Do a global search for "metrics" and you'll find the buggy function.
Round 4 is another conversation with the AI PM to get the requirements for a new feature. It's similar to round 1, but the feature is more complex than round 1's. The post above described this in a lot of detail too.
For round 5, my suggestion is to have the AI submit after implementing each piece of functionality, because you will definitely not have enough time. I ended up clicking submit at the very last minute, but the code signal just kept spinning, and by the time the grace period ended I still didn't know whether my submission had actually gone through.
Time is very tight — there's no point doing manual testing yourself, and no point analyzing the codebase either, because you simply won't have time. Honestly, I don't know what this test is even supposed to measure. Typing speed? Prompt quality?
Good luck everyone. From what I heard from the recruiter, a score in the 600s should be fine right now.
Discussion
Loading comments…