Meta Phone Screen With Two Coding Problems? Start the Second by Minute 22

The Meta phone screen is a 45-minute call. Only about 35 minutes of it is coding. In that window you usually get two problems. Meta’s own prep guide lists five minutes of introductions, 35 of coding and five for your questions. The code doesn’t run, so you test it by hand. The common way to fail is to spend most of the window on the first problem. You get a working answer. The interviewer asks for a faster one. Coding that follow-up eats the clock. Then the second problem gets eight rushed minutes and no testing. The fix is a hard switch time. Start the second problem by minute 22 of the call, whatever state the first one is in. If a follow-up comes late, answer it in words. Give the new complexity and offer to code it. Below, one Android engineer runs the screen both ways. A switch script you can adapt follows.

The Meta phone screen gives you about 35 coding minutes for two problems

Meta’s careers blog has a post on preparing for the software engineering interview. It was last updated on October 6, 2022, so check the details with your recruiter. It describes the initial screen as a coding interview with a Meta engineer. The agenda is five minutes of introductions, 35 minutes of coding and five minutes of questions. The post says you’ll solve two coding problems. A later line says “one or more”, so the count can vary.

The same post says questions are sized to solve in 10 to 30 minutes. Two problems at the top of that range don’t fit in 35. So the screen also tests whether you can stop working on a problem at the right time. The post says it directly. It tells you to manage your time. If a better solution isn’t coming, start with a working one.

Two prep sites add detail. Both sell interview coaching, so read them as interested parties. The Hello Interview guide to Meta E4 is undated. It says most candidates get a medium problem and then an easier one. It also says you can’t run your code in the CoderPad session. Kevin Landucci’s interviewing.io guide to Meta’s hiring process agrees that code execution is off. It also quotes an unnamed Meta interviewer. In that interviewer’s account, failing to solve both problems means “No Hire”. That holds even if the candidate did well on one. It’s one person’s view, not a published policy. It does fit a screen built around two problems.

Dana’s first problem works at minute 17 and her screen is lost by minute 30

Here is the running example. Dana is an Android engineer with five years of Kotlin, interviewing for a mid-level role. Meta’s post says to use the language you’re best at unless your recruiter says otherwise. So she codes in Kotlin.

Her first problem is our own example, in the usual shape of a medium. She gets a list of app launch events. Each event has a user ID. She has to return the k users with the most launches. She asks two clarifying questions and talks through a plan. By minute 17 she has this.

fun topUsers(launches: List<String>, k: Int): List<String> {
    val counts = HashMap<String, Int>()
    for (user in launches) counts[user] = (counts[user] ?: 0) + 1
    return counts.entries
        .sortedByDescending { it.value }
        .take(k)
        .map { it.key }
}

It counts in one pass and sorts the distinct users. That’s O(n + u log u), where u is the number of distinct users. It’s correct and it took 12 minutes of coding. Then the interviewer asks whether she can avoid sorting every user.

Dana knows the answer. A min-heap of size k gives O(n + u log k). But she wants to show it, so she codes it. The priority queue comparator takes a few tries without a compiler. She hand-tests it on her example and finds an off-by-one in the heap check. It’s minute 30 when she opens problem two.

Problem two is easier. It asks her to merge two sorted lists of timestamps. She writes it in seven minutes, with no time left to trace it. The loop drops the last element of the longer list. The interviewer spots it before she does. Meta’s post tells you to find and fix your own bugs. So that miss counts against her.

Nothing Dana did was wrong on its own. The heap answer was correct. The trouble was where she spent minutes 17 to 30. A working first answer plus a spoken follow-up would have bought her a clean second problem.

Code that can’t run moves minutes into testing

Without a run button, a hand trace is your only test. Meta’s post tells you to practice in a plain editor without a compiler. Our post on using Android Studio in a coding interview covers why your IDE habits won’t carry over. So every problem needs a few minutes of tracing after the code is written.

It’s easy to forget those minutes on the second problem. You plan to finish it and leave no time to test it. Dana’s merge bug is the usual result. The trace is what would have caught it. The second problem is often the easier one. So a bug there can look like carelessness instead of difficulty.

A 45-minute budget with the switch at minute 22

Here is the split Dana uses on her next attempt. The minute marks are our own suggestion, built from Meta’s agenda. Adjust them to your pace. Keep the switch time fixed.

0-5     Introductions. Keep your background to about a minute.
5-7     Problem 1: restate it, ask about input size and edge cases.
7-9     Say the approach and its complexity before writing code.
9-18    Code problem 1.
18-21   Trace it by hand on one normal case and one edge case.
21-22   Follow-up? Answer in words, give the complexity, offer to code it.
22      Start problem 2, whatever state problem 1 is in.
22-24   Restate, clarify, say the approach.
24-33   Code problem 2.
33-36   Trace problem 2 by hand.
36-40   Spare time. Code the earlier follow-up if the interviewer wants it.
40-45   Your questions.

The switch time is the only line you can’t move. If problem one isn’t working by minute 22, say what’s missing and move on. Meta’s post says “It’s better to have non-optimal but working code than just an idea.” That argues for working code on both problems before you polish either one.

The plan also keeps a spare block at minutes 36 to 40. If problem two goes well, that’s where the heap version gets coded.

What to say when the follow-up lands late

The hard moment is minute 21. You have a working answer. Now the interviewer asks for better. Here’s what Dana says on her second attempt. Adapt the details to your problem.

“Yes. Instead of sorting all the users, I’d keep a min-heap of size k. Each count goes in. If the heap grows past k, I pop the smallest. That makes it O(n plus u log k) instead of u log u. It matters when there are many users and k is small. I’m at about 17 minutes on this problem. I can code the heap now. Or I can start the second problem and come back at the end. Which would you prefer?”

The interviewer may tell you to move on, or may want the heap coded. If they want it, code it. Then shorten the clarifying step on the next problem. Either way, the interviewer makes the call with the clock in view.

Why each part of the switch script works

The first sentence answers the question. The interviewer asked whether she could do better. “Yes” plus the data structure settles it in five seconds. The rest is detail they can stop at any point.

The complexity line carries the content. The heap’s value is the change from u log u to u log k. Saying that out loud shows she understands why the heap helps. Typing out a priority queue mostly shows she knows Kotlin’s API.

The sentence about when it matters shows judgment. With few distinct users, the sort is fine. Naming the case where the heap wins shows she isn’t optimizing by reflex.

The time check and the question hand the call to the interviewer. They know the plan for the round. You don’t. Asking also avoids a silent choice that looks like poor time sense. Our post on taking a hint in a coding interview covers a related moment. The Hello Interview guide lists communication as one of four scored areas, next to testing and debugging.

Practice two problems back to back on one clock

Single-problem practice hides this trap. You solve one medium, optimize it and stop. The screen never asks you to do that. So practice in pairs. Pick one medium and one easy problem. Set a 35-minute timer and write in a plain text editor with no run button. Set a second alarm for 17 minutes in. When it goes off, say the switch script out loud, even if the first answer is perfect.

Then grade the session on three points. Did problem two start by the alarm? Did both problems get a hand trace? Did you find your own bugs before reading the answer?

Ask your recruiter about the onsite too. The interviewing.io guide says Meta began piloting an AI-enabled coding round in October 2025. In its account, that round replaces one onsite coding round for some candidates. It runs 60 minutes with a built-in assistant. It isn’t part of the phone screen. Its pacing is different. Our post on the AI-assisted coding round covers that format.

Leave a Comment