Vibe-Coding in Hiring Data Engineers: How to Spot AI Generation in Tests
When hiring a mid-level data engineer to support and develop the data platform, the team received a series of test assignments. Most solutions showed signs of AI generation: from excessive optimization to templated structures. Only one looked hand-written but suffered from poor project structure and basic errors, leading to rejection.
Roles in the process: evaluating completed assessments and conducting technical interviews. Candidates with 5+ years of experience couldn't explain their own code, citing 'recommendations from the internet'. This points to a degradation in basic skills compared to 2022–2023, when mid-levels confidently broke down Scala, Spark, and Hadoop architecture.
Types of AI-Generated Solutions
Test assignments fell into two categories:
- Vibe-coding (AI operator): Code looks perfect on the surface—no errors, with over-engineering. A typical engineer wouldn't write like that for a simple task out of laziness or pragmatism.
- Copy-paste style: Fragments from ChatGPT pasted into the project. Looks human at first glance, but detailed analysis before the interview reveals inconsistencies.
Overt markers are omitted to preserve 'professional intuition'. Even managers sometimes can't distinguish Claude Code from ChatGPT, seeing it as an advanced chatbot.
Adapted Approach to Technical Interviews
Interviews focused on the test assignment with authorship checks:
- Discussing solutions with minor condition changes (e.g., 'what to change in the input?').
- Questions on the language, Spark, and data engineering, tied to the code.
- Confirming ownership: the candidate must 'own' every line, regardless of source.
No candidate admitted to using AI. In dead-end situations, excuses like 'read it online' followed. All assessments scored 2–3/10, despite 'nice guy' factors.
Market Problems and Hiring Outcomes
The mid-level data engineer market is flooded with 'AI operators'. The new generation with 5 years of experience can't answer basic questions on distributed computing or dissect their own code. This contrasts with past years, when even juniors handled it.
In the end, they hired a candidate who opened Cursor during screen sharing. HR praised the motivation and communication. The top technically skilled Russian-speaking candidate was rejected due to lower 'likeability' and the phrase 'data engineering is the same everywhere'—interpreted as low motivation.
What's Important
- AI is not a red flag if the candidate owns the code and accounts for every line.
- Basic knowledge of Spark, Scala, and Hadoop is mandatory even for mid-levels.
- Vibe-coding shows through over-engineering and templates; copy-paste through detail inconsistencies.
- Adapt interviews: focus on condition changes and code ownership.
- The market is degrading: 100% of assessments from 5-year specialists show AI traces.
— Editorial Team
No comments yet.