
I don't care much whether someone knows the framework that came out last month. Those change every quarter. I hire for the things that don't change.

1. Solid software fundamentals
Can they design a clean API, handle async code, model a database properly and debug a distributed failure? I ask them to design an endpoint with a slow LLM call inside it and see whether they think about streaming, timeouts and retries.
2. Structured debugging
I hand them a real trace where a RAG answer was wrong. Do they check the search results, the data and the schema step by step, or do they start randomly rewriting adjectives in the prompt? The first is an engineer; the second is guessing.
3. Product sense
Before writing any code, do they ask about latency, cost per transaction, data permissions and what happens when the AI gets it wrong? People who ask those questions unprompted build systems that survive production.
Key takeaways
- 01Fundamentals outlast frameworks.
- 02Test debugging with a real broken trace.
- 03Listen for "what could go wrong?" before code.