The category split that matters
"Coding exam" covers four very different things, each with different rules:
- Practice (LeetCode, NeetCode, AlgoExpert) — no rules. AI use is normal and arguably the most efficient way to learn.
- Take-home assignments — most allow or expect normal tooling, including AI. Most have a follow-up "walk me through it" round that filters copy-paste.
- Live unproctored screens — usually unspecified. AI assistance is a gray area. Read the email; if it says "use any tools you'd normally use," you're fine.
- Proctored exams (HackerRank Proctored, university exams, AWS / GCP / Azure certs) — AI explicitly prohibited. Don't.
What companies actually say in 2026
Recruiter language has converged. The most common patterns we see in candidate-shared screens:
- "Use any tools you'd normally use day-to-day." → AI allowed.
- "You may use documentation but not LLM-based assistants." → AI explicitly disallowed.
- "This is a closed-book exercise." → AI explicitly disallowed.
- "We'd love to see your problem-solving process — use any tools." → AI allowed, but expect the follow-up to test understanding.
If it's not in writing, ask. "Are AI assistants allowed for this round?" is a perfectly normal question and a recruiter who's coy about the answer is telling you something.
Where AI legitimately wins in prep
Prep is where AI moves the needle most.
- Adversarial inputs. Ask Cloak to generate three edge cases for any function you wrote. It'll catch the off-by-one you missed.
- Pattern recognition. Solve a problem, then ask "what's the pattern-category for this and three other problems in the same family?" — that's how you build cross-problem intuition.
- Verbal critique. Talk through your solution aloud, then ask the overlay to score where the explanation was hand-wavy. Behavioral interviewers grade exactly this.
- Mock interviewer mode. Cloak Pro's interview persona asks adversarial follow-ups: "what if the inputs don't fit in memory?", "what's the failure mode under network partition?". Doing 20 of these is the closest thing to real interview pressure.
Where AI hurts prep
The fast feedback loop AI gives you is a double-edged blade. If you read the answer before you've tried, you build pattern matching on the answer, not the search process. Two weeks of that and your unassisted reps will be visibly weaker.
The fix: time-box the cold attempt. 25 minutes head-down before opening Cloak. Whatever you have at minute 25, that's your real ability. Everything after is coaching.
Specific patterns for the four categories
Practice
Use Cloak liberally. Always start cold; consult after attempt; re-do the next day cold. Track which categories you keep needing help on — that's your study list.
Take-homes
Use AI as a senior pair-programmer. Be ready to defend every architectural choice. Add code comments where you'd genuinely add them — don't try to remove the AI fingerprint, just make sure the choices are yours.
Live unproctored screens (AI allowed)
Cloak is built for exactly this. Persona on, screenshot the question, talk first, glance, answer.
Proctored exams
Don't. The downside isn't losing the cert; it's the academic / professional misconduct on your record. Cloak's Terms explicitly require you to honor exam rules.
One last thing
The companies that publish "AI allowed in interviews" are the same ones that ship AI-augmented workflows in production. If a recruiter is uncomfortable with you using the tools you'd use on the job, that's a useful signal about how the team works. Calibrate accordingly.
Get Cloak
Download from the home page. Free tier with BYOK is everything you need for prep — Pro adds the company-specific dossier and resume-grounded persona for live rounds where AI is allowed.