Interviewing Engineers in 2026: What to Actually Test For When Everyone Uses AI Coding Tools

With AI now writing 27.6% of all pull requests industry-wide, a coding interview that assumes a candidate works without AI assistance is testing a skill that increasingly doesn't reflect the actual job. Here's what's genuinely worth testing for instead.

What a Traditional Coding Interview Actually Measures Now

A whiteboard-style algorithm question, done without AI assistance, increasingly measures how well someone performs in a condition they'll rarely encounter on the job — most engineering work today happens with AI tools available. That doesn't make the skill worthless, but it's a narrower and less representative signal than it used to be.

What's Actually Worth Testing For

• Judgment about AI-generated output — give a candidate AI-generated code with a subtle bug or architectural mismatch and see if they catch it, not whether they can write the code from scratch
• System design and trade-off reasoning — AI tools are far weaker at this than at generating syntactically correct code, and it remains a genuinely differentiating human skill
• Debugging a real, messy codebase — closer to daily engineering reality than a clean, isolated algorithm problem, and harder for AI tools to shortcut
• How a candidate actually uses AI tools in practice — invite them to use their normal AI-assisted workflow during a practical exercise, and evaluate the judgment behind their prompts and acceptance decisions, not just the final output

What Not to Throw Out

Fundamentals still matter — a candidate who doesn't understand what the AI-generated code is actually doing can't catch its mistakes, can't debug it when it breaks, and can't make the judgment calls that remain genuinely human work. The goal isn't removing algorithmic or fundamentals questions entirely, it's rebalancing the interview toward the skills that actually predict success in an AI-assisted role, rather than the skills a 2015-era interview process happened to test.