went through the Google SWE loop in April 2026 while still employed (yes, interviewing on lunch breaks, yes it's exhausting). here's what the coding rounds actually looked like.
OA (online assessment, if you get one): two questions, 70 minutes, via Codility. I got one medium-level graph problem (BFS variant, not hard if you recognize the pattern) and one medium-hard string manipulation thing. timed pressure matters more than raw difficulty here. they want to see you write clean, working code, not clever one-liners.
phone screen (technical): one interviewer, Google Meet + shared doc (no autocomplete, no syntax highlighting, yes it's intentional). 45 minutes. I got a single medium leetcode-style problem with two follow-ups. the follow-ups were the real test: first one pushed me to optimize from O(n^2) to O(n log n), second one added a constraint that changed the approach entirely. if you can only solve the base case, you're probably not moving forward.
onsite coding rounds: three of them in my loop. two felt like algorithm problems (one on arrays, one on trees), one was more 'coding in the large' style where I had to extend existing code they provided. the third one is where people get surprised. it's not about knowing the algorithm, it's about producing readable, maintainable code under pressure. naming things matters. structure matters.
difficulty range: mostly medium, occasional medium-hard. I didn't see any 'hard' in the pure leetcode sense. but the medium questions had layers that a lot of people don't drill.
one thing worth knowing: they do not hint you to the optimal solution. if you're stuck in an O(n^3) approach they might ask 'is there a way to do better' once, but that's it. so know when your approach is wrong and pivot cleanly.
overall: this process is grindable. it's not a gotcha. it's just high-reps medium leetcode until the patterns are automatic.