Writing code
Generating functions, tests and boilerplate. The risk is code that passes the happy path and fails in production.
Most engineers now write code with an AI assistant. Speed is no longer the signal. What matters is whether they read what the AI wrote, test the edge cases it skipped and notice when a suggestion is plausible, compiles and is still wrong.
Where software engineers use AI every day, and where it goes wrong.
Generating functions, tests and boilerplate. The risk is code that passes the happy path and fails in production.
Explaining unfamiliar code and proposing fixes. AI often fixes the symptom, not the cause.
Exploring approaches and libraries. Strong engineers check versions, licences and whether the API really exists.
The habits that separate a capable software engineer from one who copies the answer.
Ask for a specific example each time. Vague answers are a signal in themselves.
Questions show what candidates say. A task shows what they do when the AI is confidently wrong.
A payments endpoint occasionally charges customers twice. The AI colleague proposes adding a retry with a longer timeout and says the issue is network latency.
Retries make it worse: the endpoint is not idempotent, so a slow response followed by a retry creates a second charge.
Asks what happens when the first call succeeds slowly, spots the missing idempotency key and proposes it instead of the retry.
Signs a candidate depends on AI rather than working with it.
These tools are free. Our main work is measuring how well people work with AI, so employers can hire on evidence and teams can build the skill. Hiring? See AI interview questions by role.
How well a person works with AI on real work, measured across Judgment, Reasoning, Critical thinking, Questioning and Adaptability.
What AI capability isCandidates work a real task from the role alongside Marcus, an AI colleague who is sometimes wrong. You see who checked, who questioned and who just accepted.
Try the hiring demoA verifiable score for how you work with AI, with the evidence behind every rating. Teams use it to measure and build AI capability. Your first one is free.
How AIQ is measuredUsing AI assistants to write, test and review code while reading what they produce, testing edge cases, verifying suggested APIs and owning the design decisions.
Let them use AI on a realistic task where the assistant's suggestion is subtly wrong, and observe whether they test and question it. That shows the skill directly.
Puzzle-style tests lose value because AI solves them. Work-sample tests that include a flawed AI suggestion remain useful because they measure judgment.
No. It adds evidence of how a candidate works with AI on a real task, alongside your technical and team interviews.
Each candidate works a task from your role alongside an AI colleague. You get an AIQ™ score with the evidence behind every rating.