🚀 Hiring AI-First Engineers: Can They Really Code With AI?

🚀 Hiring AI-First Engineers: Can They Really Code With AI?

I’m advising a startup on its tech setup with a non technical founder. AI came up early and we pushed acceleration hard. The founder is now building vibe-coded tools, creating 𝐬𝐦𝐚𝐥𝐥 𝐏𝐑𝐬 like front-end changes, and wants to go further. To prove the point, I delivered a full feature with a third-party integration — 𝟏𝟎𝟎% 𝐀𝐈-𝐠𝐞𝐧𝐞𝐫𝐚𝐭𝐞𝐝 — in a legacy codebase and an unfamiliar language. One person, end-to-end. Faster and cheaper than me designing a solution and handing it to a team, I''d done this before so just did it.

But here’s the hitch: he wants to hire senior 𝐀𝐈-𝐟𝐢𝐫𝐬𝐭 𝐞𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐬, he's ahead of them now, candidates fail at the first AI use questions. So what should a candidate actually know? What skills should they show?

🧠 𝐖𝐡𝐚𝐭 𝐚𝐧 𝐀𝐈-𝐅𝐢𝐫𝐬𝐭 𝐄𝐧𝐠𝐢𝐧𝐞𝐞𝐫 𝐒𝐡𝐨𝐮𝐥𝐝 𝐊𝐧𝐨𝐰

* ⚡ Choose and justify a 𝐦𝐨𝐝𝐞𝐥. We want to understand their rationale here. * 🛠️ Use an 𝐀𝐈 𝐜𝐨𝐝𝐞 𝐞𝐝𝐢𝐭𝐨𝐫/𝐚𝐠𝐞𝐧𝐭 (Cursor, Claude, etc.) and explain their choice. * 📚 𝐆𝐫𝐨𝐮𝐧𝐝 𝐀𝐈 with repo context, ticket details, and non-functional requirements. * 🔄 𝐈𝐭𝐞𝐫𝐚𝐭𝐞 𝐩𝐫𝐨𝐦𝐩𝐭𝐬: refine, review, adjust not one mega prompt and wait, good prompts iteratevely applied are key, reviewing output as they go. * 🧪 Always cover 𝐧𝐨𝐧-𝐟𝐮𝐧𝐜𝐭𝐢𝐨𝐧𝐚𝐥𝐬 (tests, security, performance, observability) and manage scope. * 👀 Spot and fix 𝐛𝐚𝐝 𝐀𝐈 𝐨𝐮𝐭𝐩𝐮𝐭 — stay in control, don’t just review what AI hands back and be AIs assistant.

📝 𝐓𝐡𝐞 𝐈𝐧𝐭𝐞𝐫𝐯𝐢𝐞𝐰 𝐈’𝐝 𝐑𝐮𝐧

➊ Share a repo + Jira ticket on the day, something doable by someone new without AI in a day: afew hours of analysis, proper fix with rework, not just a text change. A good attempt should take under an hour with AI.

➋ Candidate analyses repo with AI, identifies functionality and components.

➌ Explains grounding strategy, defines 𝐍𝐅𝐑𝐬, scopes the fix.

➍ Use AI to implement, keeping scope tight. Add unit tests, or API endpoint tests if none exist for example.

➎ Open a 𝐏𝐑: If its in an 𝐀𝐏𝐈 (I think a good case), expect to discuss end-to-end integration, or manual QA steps if automation isn’t feasible.

➏ Walk through how they prompted, iterated, and reviewed and end result.

⚠️ 𝐑𝐞𝐝 𝐅𝐥𝐚𝐠𝐬

* 𝐌𝐚𝐧𝐮𝐚𝐥 𝐜𝐨𝐝𝐢𝐧𝐠. * No grasp of 𝐦𝐨𝐝𝐞𝐥𝐬 or 𝐭𝐨𝐨𝐥 options. * No 𝐭𝐞𝐬𝐭𝐬, no 𝐍𝐅𝐑𝐬, just functional AI code under pressure. * Blindly accepting 𝐀𝐈 𝐨𝐮𝐭𝐩𝐮𝐭 without iteration. * 𝐏𝐨𝐨𝐫 𝐏𝐑 quality — if it’s weak, that’s on the engineer not the AI.

💡 𝐅𝐢𝐧𝐚𝐥 𝐓𝐡𝐨𝐮𝐠𝐡𝐭: AI makes one engineer a 𝐟𝐨𝐫𝐜𝐞 𝐦𝐮𝐥𝐭𝐢𝐩𝐥𝐢𝐞𝐫 — I’ve done it more than once. But to hire for that, you need to test for 𝐬𝐤𝐢𝐥𝐥𝐬, 𝐣𝐮𝐝𝐠𝐦𝐞𝐧𝐭, 𝐚𝐧𝐝 𝐝𝐢𝐬𝐜𝐢𝐩𝐥𝐢𝐧𝐞, not coding skills.

👉 𝐖𝐨𝐮𝐥𝐝 𝐲𝐨𝐮𝐫 𝐞𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐬 𝐩𝐚𝐬𝐬 𝐭𝐡𝐢𝐬 𝐭𝐞𝐬𝐭, 𝐨𝐫 𝐟𝐚𝐢𝐥 𝐢𝐭?

#AIEngineering #Hiring #Interviewing #DeveloperExperience