Most interview loops are long because nobody's confident about what to listen for, so they add more rounds hoping something surfaces. It's the wrong fix. A well-structured 30-minute conversation, aimed at the right four signals, tells you more than six hours of unfocused conversation ever will. The length of a loop isn't a proxy for its rigor. The specificity of what you're listening for is.
Why a short conversation can be enough
Long loops exist to catch signal that a shorter, unfocused conversation might miss, but the length is compensating for a lack of aim, not adding real rigor. If you know exactly which four things predict on-the-job performance, you can get all four in half an hour by asking pointed questions and following up hard on the answers. What makes a 30-minute interview weak isn't its length, it's asking generic questions and accepting generic answers. What makes it strong is asking for specifics and refusing to move on until you get them.
Signal one: how they talk about a real past decision
Ask for one specific decision they made on a real project, ideally one with a genuine trade-off, not a success story. Listen for whether they can describe the actual alternatives they considered, what they gave up by choosing what they chose, and what they'd do differently now. Candidates who've genuinely done the work describe the mess: the option that looked good on paper and wasn't, the constraint nobody warned them about. Candidates reciting a rehearsed accomplishment describe a clean, linear story with no real trade-off in it at all.
- Ask: "Tell me about one technical decision on a real project where you weren't sure you were right."
- Follow up: "What was the alternative you didn't pick, and what did that cost you?"
- Follow up: "If you made that decision again today, what would you change?"
- A real answer has friction in it. A rehearsed answer sounds like a highlight reel.
Signal two: how they reason about trusting AI output
This is one of the highest-value signals available today, and one of the least commonly asked about. Anyone can generate output with an AI tool now; what separates a strong practitioner is knowing when to trust it and when to verify it by hand. Ask about a specific instance: a time an AI tool's output looked right and wasn't, or a time they had to decide how much to trust a generated result under time pressure. Strong candidates describe a concrete method, specific checks, specific red flags they watch for, not a general philosophy about 'always double-checking.'
| Question | Weak answer | Strong answer |
|---|---|---|
| When do you trust AI-generated code or output? | "I always review everything carefully" | Names specific categories they verify (edge cases, external data, security-sensitive logic) and specific ones they don't re-derive by hand |
| Tell me about a time generated output was wrong | Vague or can't recall a real instance | Describes the specific output, what tipped them off, and what they changed as a result |
Signal three: specificity versus vagueness about their own work
Ask a candidate to describe a project they built, then ask three specific follow-ups: what exactly broke during it, what the actual numbers were (latency, accuracy, cost, whatever's relevant), and what they'd do differently with what they know now. People who did the work have these details on hand without needing to think hard, because they lived them. People describing work they were adjacent to, or work they're overstating, tend to answer in generalities and get less specific, not more, as you push for detail.
Signal four: how they respond to being challenged
Push back, respectfully but genuinely, on something they've just said. Disagree with a technical choice they described, or ask why they didn't consider an alternative approach. Strong candidates engage with the substance: they either defend the choice with real reasoning or update their view when the pushback is fair. Weak signal shows up as either collapsing immediately without engaging, or defending the original answer no matter what's actually being raised. Both are the same underlying problem: no real reasoning under the answer, just the answer itself.
- Disagree with one specific choice they described, and see whether they engage with the reasoning or just repeat their answer.
- A candidate who updates their view when the pushback is genuinely fair is a stronger signal than one who never budges.
- A candidate who collapses instantly, with no defense of a choice they clearly thought through, is a signal too, just a different one.
