Design rounds that go deep, and behavioral rounds that probe scope. Practise for the band where architecture and influence are the signal.
At SDE-3 the coding round stops being the discriminator. Everyone in the loop can code. What decides the outcome is whether your system design holds up when an interviewer who has built the real thing starts pulling on it, and whether your behavioral stories show scope beyond your own commits.
Expect design rounds that deliberately go past your prepared depth. The interviewer will pick the component you glossed over and stay there: how does the write path stay consistent during a partition, what happens to in-flight jobs during a deploy, what is your migration plan for the schema change you just proposed. Saying "I would use Kafka" without explaining delivery semantics reads as pattern matching, not experience.
The behavioral round at this band is a scope interview in disguise. You need stories where you set the technical direction, aligned people who did not report to you, and owned the outcome including the part that went wrong.
45 min
One or two rounds, usually a filter rather than the deciding signal.
60 min
Open-ended design with sustained follow-up on the hardest component.
60 min
Migration strategy, failure modes, operational cost, and what you would not build.
45-60 min
Cross-team influence, mentoring, and decisions you owned that were reversed.
45 min
Scope calibration - what you actually drove versus what your team drove.
5-9 years
Design depth, leadership
6-10 years
Architecture, ownership
5-8 years
Design, cross-team impact
8+ years
Strategy, technical direction
The SDE-2 round rewards a complete, reasonable design. The SDE-3 round assumes that and spends most of its time on one hard sub-problem. You will be asked about consistency during failure, migration without downtime, and operational cost - areas where reading blog posts does not substitute for having run the system.
Enough not to fail the round. Most SDE-3 loops still include one coding interview, and being rusty there can sink an otherwise strong candidate. But marginal hours are better spent on design depth and behavioral scope, which is where the decision is actually made.
Scope and ownership. The story needs a decision you made, the constraint that made it hard, people outside your reporting line you had to bring along, and a measured outcome. Stories where you executed someone else’s plan well read as SDE-2 signal no matter how senior you are.
This is the band where it pays off most. An AI interviewer gives unlimited reps, but calibration against the senior bar - is my scope actually L5 - needs someone who has sat on the hiring side of those loops.
Run a deep design mock, then validate your level with a 1:1 expert session.