Can you defend the AI-assisted work sample?
A fictional startup gives you a 90-minute exercise. AI use is allowed, but you must understand and explain your work.
The question
This is an original RelayDesk practice task, not a real company’s interview exercise. The supplied rules allow AI for this whole task. A fixed test set contains 20 cases. Acceptance requires at least 95% passing and zero cases with lost information.
| Cases | |
|---|---|
| Pass | 16 |
| Format failure | 2 |
| Information lost | 2 |
The three groups are disjoint. You spent 30 minutes drafting and 20 minutes running and reviewing these checks.
“Choose your next action and explain the trade-off.”
Not saved by this site. Your browser may restore text when you navigate back.
A strong answer, step by step
Open each step in turn. This is one strong answer, not the only one.
Step 1: Can this version be submitted?
What would you do here? Open the move you’d make.
Check both acceptance conditions
The strong move. A polished output cannot replace the stated checks.
Submit it because most examples look right.
Tempting, but it misses something. The fixed set already shows failures, including lost information.
See how a strong candidate says it
“It passes 16 of 20 cases, or 80%, and loses information in two cases. It fails both acceptance conditions.”
Why it works: A polished output cannot replace the stated checks.
Step 2: What do you fix first?
What would you do here? Open the move you’d make.
Repair the lost information
The strong move. Lost information is the higher-impact failure and is explicitly disallowed.
Spend the remaining time improving the demo video.
Tempting, but it misses something. A stronger demo leaves the known information loss unresolved.
See how a strong candidate says it
“Find why information is lost and correct that behavior before polishing presentation.”
Why it works: Lost information is the higher-impact failure and is explicitly disallowed.
Step 3: How much time remains?
Repair and repeat checks take 12 minutes in total. The repeat checks now pass 18 cases; two format failures remain and no case loses information.
What would you do here? Open the move you’d make.
Recheck quality and the time budget
The strong move. The repair clears one gate, not both.
Call it finished because the serious failures are gone.
Tempting, but it misses something. The pass-rate requirement remains unmet.
See how a strong candidate says it
“After the 12-minute repair, 28 minutes remain. The retested result is 18 of 20, or 90%, with no information loss; it still misses 95%.”
Why it works: The repair clears one gate, not both.
Step 4: What can you complete and explain?
You estimate 6 minutes for the next format correction and 8 minutes for a full repeat check. No result from that work exists yet.
What would you do here? Open the move you’d make.
Fix, rerun and explain
The strong move. A planned fix is not a passing result. The submission must include observed checks and your reasoning.
Claim the fix will pass and use all remaining time on presentation.
Tempting, but it misses something. That replaces observed evidence with a forecast.
See how a strong candidate says it
“Use 6 minutes for the format correction and 8 for the full repeat check, leaving 14 minutes to explain the work. Claim 95% only if at least 19 cases actually pass.”
Why it works: A planned fix is not a passing result. The submission must include observed checks and your reasoning.
The answer, in short
- Decision
- Submit only the observed result, with the checks and decisions you can explain.
- Because
- The latest verified pass rate is 90%; the next fix is still a plan.
- What would change it
- A completed repeat check could establish at least 19 of 20 passing with zero information loss.
Common traps
- Do not claim a predicted fix as a test result.
- AI permission does not remove responsibility for understanding.
Check your answer
- I used the exercise’s own AI rule.
- I checked both quality gates.
- I counted verification time.
- I separated observed results from planned fixes.
Last reviewed Nov 10, 2026