The three AI regimes and their rules

Each round sets its own AI rule: confirm it, follow it exactly, and make allowed AI use visible.

Before you read: predict

Your Meta onsite has two coding rounds, and the recruiter's schedule labels the second one "AI-enabled coding". It opens on a multi-file project with a failing test. The interviewer says, "Find what's broken and fix it," and nothing about the assistant panel on the right.

As of 2026-10-01, going by Hello Interview's report of Meta's format, which statement about using the assistant at this moment holds?

The 30-second version

Interview rounds fall into three AI regimes: banned, one AI-enabled round beside classic ones, and AI-native rounds that expect an assistant. Teams are split, so you need both modes. A rule belongs to a round, sometimes to one phase, and formats change. Confirm it each time, follow it exactly, and where AI is allowed, say what it gave you.

Each round sets its own AI rule: confirm it, follow it exactly, and make allowed AI use visible.

How it works

Step 1 of 10

Banned. A classic round allows no assistant. In one 2026 survey, 34% of hiring teams banned AI in interviews entirely. [S1]

Analogy: A pub quiz whose quizmaster sets the phone rule round by round

At a pub quiz, phones stay face down in most rounds, and a phone on the table voids the team's round. One night has a single "look it up" round where phones are allowed. A treasure-hunt night is all searching: a team that leaves its phones in its bags gives up the point of the night. The quizmaster can change a rule as a round starts, so a captain who is unsure asks before the round, and a team that looked an answer up says so on the sheet.

At the pub quiz	In the interview loop
A phones-away round	A banned, classic round
One look-it-up round on an ordinary night	One AI-enabled round beside classic ones
A treasure-hunt night	An AI-native round
The quizmaster changing the rule as a round starts	A phase rule set by the interviewer
"Looked up" written on the sheet	Saying the code came from the assistant

Breaks down when: a quiz scores only the answer, while an interviewer scores how you reached it. Wrong inference: "in an AI round, a correct answer from the assistant is the whole score." Quiz rules are also stable for years; interview formats changed within months.

In real terms: before each round, and each phase, know its AI rule and its source.

Examples, run for real

EXH-2220 is a synthetic log of a practice onsite for an ExamHall frontend role, with three rounds. The candidate opened the assistant once in each round. The re-run, after the rules were collected and written down, is MR !2520.

EXH-2220 mock onsite: 3 rounds · rules stated 2 · assistant uses 3
  R1 classic coding · rule "no AI" (recruiter, 2026-09-14) · API lookup -> breaks the stated rule
  R2 take-home · rule "AI allowed; say where" (brief, 2026-09-14) · generated tests, README silent -> use not disclosed
  R3 system design · rule never asked · diagram draft -> used under an unconfirmed rule
  flags: 3
MR !2520 re-run: 3 rounds · rules stated 3 · assistant uses 1
  R2 take-home · rule "AI allowed; say where" (brief, 2026-09-14) · generated tests, README names the tool and the tests -> covered
  R3 system design · rule "no AI" (recruiter reply, 2026-09-18)
  flags: 0

In the first run, each use breaks a rule in a different way. R1's rule was clear and the lookup ignored it. R2 allowed the assistant, but nothing told the reviewer where it had been used. R3's rule was never asked for, so the candidate guessed, and the recruiter's later reply showed the guess was wrong.

In the re-run, the candidate asked about every round before the loop and wrote each answer down with its source and date. R1 and R3 had no assistant, and R2's README said which tests the assistant drafted. The checker reports nothing to fix.

The script compares a log against stated rules. It cannot tell whether a rule was reported correctly, and the log and rules are the author's fixture, not a real loop.

Try it

Answer first, then check. Nothing is saved.

Question 1 of 5

A learner has loops booked at four companies and plans to rehearse checkout-cart tasks with an assistant switched on throughout, because "serious teams allow AI now." Which statement about interview AI rules, as published by 2026-10-01, bears on that plan?

Question 2 of 5

A frontend candidate reaches Canva's AI-Assisted Coding interview, as it was described on 2026-10-01. To show strong fundamentals, she keeps the assistant closed and hand-writes the checkout discount engine.

How does that choice sit with the round?

Question 3 of 5

In the build phase of Meta's AI-enabled coding round, with the assistant in use, a candidate asks it for a helper that rounds each checkout line's tax and pastes it into cart.ts. The interviewer has been watching the editor, not the chat panel.

As of 2026-10-01, going by Hello Interview's guide, which line, said at the paste, fits the round?

Question 4 of 5

In September 2026 a learner preparing for a Meta software-engineer onsite plans from notes she saved in early 2025: "Meta coding rounds: two, LeetCode-style, no AI." She will rehearse both coding rounds, including a checkout-totals task, with the assistant off.

What should she change?

Question 5 of 5

Two weeks before Meta's AI-enabled coding round, a learner moves all prep time to prompt templates and stops the practice she did for classic rounds, reasoning that "this round grades how I use AI." Her practice task is a checkout cart.

Which change to her plan fits how the source says that round is scored, as of 2026-10-01?

Go deeper

Pro tip

When a recruiter tells you a round's AI rule on a call, reply in writing with the rule as you understood it, so a misunderstanding surfaces before the day.

Checked against the tools and docs as of 2026-10-01.

Predict first