Skip to content
AI Engineer Guide.

concept · career · intermediate · concept 4 of 33

Interview formats

0 of 33 complete

Why it matters

A loop is designed to sample different abilities, and candidates who prepare one way for all of them underperform in the rounds that do not match their habit. A strong system design answer does not help in a debugging round where the interviewer wants a hypothesis and a measurement. Knowing the format tells you what to optimise: structure for design, narration for coding, evidence for behavioural, and a written artefact for take-homes.

Mental model

A loop is a portfolio of measurements, each with its own instrument. The metaphor breaks because the instruments are people forming impressions, so the same answer can score differently depending on whether the reasoning was visible to the person holding the instrument.

How it works

Each stage scores a different behaviourInteractive
startquestionoutcomepath takenruled out
VerdictChoose· recommended hereConditional· with the named cost

Each branch names the stage and the behaviour it scores; a prepared answer does not transfer between rounds.

Step 1 of 6

Ask for the stage list

Write the stage list, duration and tools down before preparing. The format decides what to optimise: structure, narration, evidence or an artefact.

start

Which stage are you preparing for?

Ask the recruiter for the stage list, the duration and the tools allowed, and write them down before preparing. Each stage samples a different behaviour.

Path walked

  1. Which stage are you preparing for?
Step 1/6 · Ask for the stage list

Figure. Each stage samples a different ability: narration for coding, structure for design, a hypothesis for applied debugging, evidence for behavioural, and a written artefact for take-homes. The branch names what to optimise, not what to know.

What to notice. The mapping is the page's, not a scorecard; stage lists differ by company and the recruiter's list wins. No duration or weight is drawn because the page names none; “often weighted less” is the page's description, not a measured ratio. The figure does not claim a take-home can be skipped — it claims one stage's preparation does not cover another.

Text version

A decision map from the stage list to what each stage scores. The question asks which stage you are preparing for. The coding branch asks for narration: narrate the decision, not the syntax, because the fundamentals still decide. The design branch asks for requirements, request path, failure and cost, in that order. The applied debugging branch asks for a hypothesis and a measurement before touching the code, because design preparation does not transfer. The behavioural branch asks for three stories with decisions and evidence rather than a list of achievements. The take-home branch is conditional: a written artefact, one signal among several and often weighted below the onsite rounds. Six steps walk the branches, starting with the stage list.

  1. Ask the recruiter for the stage list, the duration and the tools allowed, and write it down before preparing.
  2. Map each stage to the skill it samples, then check which of those skills you have never demonstrated aloud.
  3. For applied rounds, practise reading a broken system and stating a hypothesis before touching code.
  4. For design rounds, practise requirements, a request path, a failure mode and a cost, in that order.
  5. For behavioural rounds, prepare three stories with decisions and evidence rather than a list of achievements.
  6. After each real interview, write the questions you were asked and the answer you gave while it is fresh.

Trade-offs

ChoiceGood forCostFails when
Broad preparationCovers every stageNo stage gets deep practiceThe loop weights one stage heavily
Depth in one areaStrong signal where you are bestA weak stage can veto the loopThe role requires breadth
Practice with peersRealistic pressure and feedbackScheduling and mismatched levelsFeedback is polite rather than specific
Solo rehearsalCheap and repeatableNo interruption practiceThe round is interactive and you have never been interrupted
Trade-off mapInteractive

Good for

Covers every stage

Cost

No stage gets deep practice

Fails when

The loop weights one stage heavily

  1. The loop weights one stage heavily
  2. The role requires breadth
  3. Feedback is polite rather than specific
  4. The round is interactive and you have never been interrupted

Figure. Each choice with what it buys, what it costs, and the case where it loses.

What to notice. The fourth column is the decision: if the losing case describes your situation, the choice is wrong.

Text version

Four trade-offs: Broad preparation is good for Covers every stage, costs No stage gets deep practice, and fails when The loop weights one stage heavily. Depth in one area is good for Strong signal where you are best, costs A weak stage can veto the loop, and fails when The role requires breadth. Practice with peers is good for Realistic pressure and feedback, costs Scheduling and mismatched levels, and fails when Feedback is polite rather than specific. Solo rehearsal is good for Cheap and repeatable, costs No interruption practice, and fails when The round is interactive and you have never been interrupted

Failure modes

  • Preparing content only. You know the material and freeze on the format. Debug: time one answer per format and check whether it fits the stated duration.
  • Ignoring the applied round. Design preparation does not transfer to debugging a failing system. Debug: rehearse one broken-system scenario end to end and note where you started.
  • No record of the loop. The same mistake repeats across companies because nothing was written down. Debug: check whether you can list the last five questions you were asked.

In production

Keep a running log of questions, formats and your answers, and review it before the next loop. Rehearse under time pressure at least once per format.

Common misconceptions

People think: AI interviews are all about model knowledge.

Actually: Loops still weight coding, design and collaboration; domain knowledge is one input among several.

People think: The take-home is the deciding round.

Actually: It is one signal, often weighted less than the onsite rounds, and a strong take-home cannot compensate for a weak debrief.

Interview angle

Asked how you prepare for a loop, the strong candidate names each stage, the skill it samples and the evidence they will bring to it, then says what they would change after a weak round. Weak answers treat every round as a knowledge quiz and never mention the format's mechanics.

Sources