The art of interviewing in the age of LLMs. Why technical skills are taking a backseat to logical intent.
For the past two decades, technical interviews have been a ritual of syntax recall, data structure traversal, and whiteboard algorithms. The implicit assumption: if you can code a binary search tree from memory, you can build production software.
That assumption is now obsolete.
The LLM Inversion
When every engineer has access to a co-pilot that can write a merge sort, implement a REST endpoint, or debug a race condition in seconds, the value of memorized solutions collapses. What doesn't collapse — what actually becomes more valuable — is the ability to decompose problems, specify requirements precisely, and evaluate generated output critically.
In other words: the skill of prompting well.
What to Interview For Now
The most effective technical interviews we've seen in 2024 look less like coding challenges and more like design sessions:
- Give the candidate a vague, real-world problem and watch how they clarify requirements.
- Provide them with an LLM and evaluate not what they build, but how they direct the build.
- Ask them to review AI-generated code for security, performance, and correctness — not write it themselves.
The engineers who excel in this format aren't necessarily the strongest coders. They're the strongest thinkers — the ones who understand systems deeply enough to know when a generated solution is subtly wrong.
The Uncomfortable Implication
If we're honest, this shift devalues a certain type of interview performance that has long been correlated with prestige: the candidate who can implement a red-black tree without reference material. That skill has never been a strong predictor of on-the-job performance. We just acted like it was.
The LLM era is giving us an opportunity to fix that. The question is whether we're willing to redesign our processes or keep running interviews optimized for a world that no longer exists.