AI hasn't replaced engineers — it's raised the bar. Practice on deliberately vague problems. Get scored on how well you think, not just what you ask.
Your Prompt
A few guesses — frontend is probably re-rendering too much, the API gateway might be hitting a distant region for some users, and the queries likely aren't indexed. I'd start with the query layer first.
AI Response
Good instinct to narrow it down. It's slow for everyone on the main feed — and yes, the DB is the culprit. Queries aren't indexed and the feed hits it on every load...
The shift
For decades, the ability to write code was the scarce skill. Now AI can write the code. What it can't do is think clearly about the problem.
The new leverage point is upstream: defining constraints, surfacing tradeoffs, asking the right questions. In other words — prompting well.
Before AI
Write the implementation
Now
Define the requirements precisely
Before AI
Know every API
Now
Know what to ask for
Before AI
Debug line by line
Now
Narrow the problem space quickly
How it works
Every problem is deliberately under-specified — just like real engineering work. No hand-holding.
Ask clarifying questions, define constraints, explore tradeoffs. The AI mirrors the quality of your thinking.
Your prompts are evaluated — not the AI's answers. Better questions unlock better scores and deeper insights.
Community
First thing I'd lock down is auth — who can shorten a URL, and can they see analytics for links they didn't create? Permissions need to be airtight before we talk scale.
At 10K redirects per second the bottleneck is the read path. I'd assume a CDN-edge cache in front of everything and work backwards from there. Write path barely matters.
Before any infra — what does a broken redirect cost us? If users hit a dead link once they churn. I'd rather over-engineer reliability than chase p99 latency on day one.