So it seems like Stripe recently added a new AI Programming Exercise round to their onsite loop.

Sharing what we’ve gathered from a few candidates who went through it recently:

The interview runs in HackerRank, but it’s not the standard coding setup. There’s a built-in AI chat window, kind of like a lightweight Cursor, that you can talk to. You can ask it to read the README, come up with a plan, write code, add tests, debug, and so on.

The task itself looks roughly like this:

You’re given a list of transactions and a list of rules. Each rule says whether to accept or block a transaction, followed by an if condition. You need to parse the rules and decide whether a transaction matches the condition.

The problem consists of multiple parts. It starts pretty straightforward, mostly keyword and string matching, but later parts get trickier with boolean logic like AND / OR, and each section builds on the previous ones. There’s a detailed README, so reading and understanding the spec quickly is a big part of it.

The actual coding portion is apparently only around 30 minutes, so trying to hand-code everything yourself may not be realistic. The expectation seems to be that you lean heavily on the AI. You can technically write it all yourself, but there’s a good chance you won’t finish.

A strategy that worked for people:

1. Have the AI read the full README.
2. Ask it to summarize the requirements.
3. Get a proposed implementation plan and actually review it.
4. Let it write the code.
5. Add your own tests and edge cases.
6. Run it, debug, and understand everything.

Our impression is that this round is testing whether you can use AI effectively without turning your brain off.

The AI can handle a lot, but it may over-engineer, miss edge cases, or make assumptions that aren’t actually in the README. So the key seems to be: Understand the spec, guide the AI, review the generated code, write meaningful tests, catch issues, and explain your reasoning under time pressure.