{"id":779,"date":"2026-10-20T09:40:00","date_gmt":"2026-10-20T07:40:00","guid":{"rendered":"https:\/\/grindloop.io\/blog\/?p=779"},"modified":"2026-10-17T15:35:22","modified_gmt":"2026-10-17T13:35:22","slug":"in-person-coding-interview-hand-trace","status":"publish","type":"post","link":"https:\/\/grindloop.ai\/blog\/in-person-coding-interview-hand-trace\/","title":{"rendered":"In-Person Coding Interview? Your Hand Trace Replaces the Run Button"},"content":{"rendered":"<p>An in-person coding interview takes away the one tool most Android engineers lean on hardest. That tool is the run button. On a whiteboard or on paper, nothing compiles and no test goes green. So the interviewer watches how you check code you can&#8217;t execute. Google is the clearest case. Sundar Pichai said in June 2025 that Google would add at least one in-person round. He said its purpose was to make sure the fundamentals are there. In our view, the first fundamental a whiteboard tests is reading your own code without a machine. The tactic is a hand trace. Write the solution. Then walk one small input through it line by line, out loud. Track every variable in a table next to the code. Then trace an edge case like an empty list. Done well, the trace finds your bug before the interviewer does. Below is one merge-intervals problem, with two bugs in the first draft, traced on a board.<\/p>\n<h2>The in-person round came back to check what AI can&#8217;t fake in the room<\/h2>\n<p>At Google, the change began with employee questions. CNBC <a href=\"https:\/\/www.cnbc.com\/2025\/03\/09\/google-ai-interview-coder-cheat.html\" target=\"_blank\" rel=\"noopener\">reported in March 2025<\/a> on a February town hall, based on an audio recording it reviewed. Employees asked whether onsite interviews could come back. Pichai answered that some fraction of interviews being in person was worth considering. Brian Ong, Google&#8217;s vice president of recruiting, said interviewers are told to probe candidates on their answers. Probing checks whether you understand what you wrote.<\/p>\n<p>By June 2025 the idea was firmer. Lex Fridman asked Pichai how Google&#8217;s coding interviews had changed with AI. He said Google would add at least one in-person round. The goal, in his words, was &#8220;just to make sure the fundamentals are there.&#8221; That&#8217;s from the <a href=\"https:\/\/lexfridman.com\/sundar-pichai-transcript\/\" target=\"_blank\" rel=\"noopener\">podcast transcript<\/a>. In the same answer, he called skill with AI coding tools an asset. So in his telling, Google still values AI skill. The in-person round checks the fundamentals underneath it.<\/p>\n<p>Google isn&#8217;t alone. Computerworld <a href=\"https:\/\/www.computerworld.com\/article\/4044734\/to-counter-ai-cheating-companies-bring-back-in-person-job-interviews.html\" target=\"_blank\" rel=\"noopener\">wrote in August 2025<\/a> that Cisco and McKinsey had also moved interviews back in person. Google&#8217;s remote rules point the same way. Its <a href=\"https:\/\/services.google.com\/fh\/files\/misc\/general_virtual_interviews_candidate_resource.pdf\" target=\"_blank\" rel=\"noopener\">candidate guide for virtual interviews<\/a> is undated. It says using AI during the interview leads to disqualification. It also suggests a thick marker and paper as a backup for sharing code over video.<\/p>\n<h2>Without a run button, the interviewer watches how you check your code<\/h2>\n<p>In a remote round, many candidates find bugs by running the code. A failing test points at the problem, so you fix it and run again. Our post on the <a href=\"https:\/\/grindloop.ai\/blog\/codesignal-general-coding-assessment-score\/\">CodeSignal General Coding Assessment<\/a> covers a milder version of this. There you lose Android Studio but keep a compiler and tests. In person, you lose execution entirely.<\/p>\n<p>That changes what a bug costs. Remotely, a missed edge case fails a test. You fix it in a minute. On a whiteboard, it stays hidden until someone reads the code carefully. If the interviewer finds it first, the note can say you didn&#8217;t verify your own work. If you find it first, the note can say you test without tools.<\/p>\n<p>So practice checking code by hand. A misremembered method name is easy to correct out loud. A wrong answer you never noticed stays in the notes.<\/p>\n<h2>The running example: merge overlapping screen sessions<\/h2>\n<p>Say the prompt is a classic with an app framing. You get a list of sessions, each a start and end minute. A user can have overlapping sessions across screens. Return the merged sessions, so the total time on the app is easy to sum.<\/p>\n<p>Ask one clarifying question first. If one session ends at minute 10 and another starts at 10, is that one session? Assume the interviewer says yes. Then here&#8217;s a first draft, written the way it often comes out under pressure.<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"\">data class Session(val start: Int, val end: Int)\n\nfun merge(sessions: List&lt;Session&gt;): List&lt;Session&gt; {\n    val sorted = sessions.sortedBy { it.start }\n    val result = mutableListOf&lt;Session&gt;()\n    var current = sorted[0]\n    for (s in sorted.drop(1)) {\n        if (s.start &lt;= current.end) {\n            current = Session(current.start, maxOf(current.end, s.end))\n        } else {\n            result.add(current)\n            current = s\n        }\n    }\n    return result\n}<\/pre>\n<p>It reads as correct. It sorts by start and extends the current session on overlap. When a gap appears, it pushes the session. The <code>&lt;=<\/code> covers the touching case the interviewer confirmed. Two bugs remain. In an IDE, the first test run would expose one of them. On a board, only a trace will.<\/p>\n<h2>A hand trace on two inputs finds both bugs<\/h2>\n<p>Pick a small input that hits each branch. Three sessions do it. Two of them overlap. The third comes after a gap. Write them out of order, so the sort matters too. Then draw a table beside the code and fill one row per loop step.<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"\">input:  (20,25) (0,10) (5,12)\nsorted: (0,10) (5,12) (20,25)\n\nstep  s        branch    current   result\ninit  -        -         (0,10)    []\n1     (5,12)   overlap   (0,12)    []\n2     (20,25)  gap       (20,25)   [(0,12)]\nend   -        return    (20,25)   [(0,12)]   &lt;- (20,25) is lost<\/pre>\n<p>The last row shows the first bug. The final session is held in <code>current<\/code> when the loop ends. Nothing adds it to <code>result<\/code>. The fix is one line after the loop, <code>result.add(current)<\/code>.<\/p>\n<p>Then trace the edge case, an empty list. The trace stops at the first line that uses data. <code>sorted[0]<\/code> on an empty list throws an <code>IndexOutOfBoundsException<\/code>. The fix is a guard at the top, <code>if (sessions.isEmpty()) return emptyList()<\/code>. We ran both versions with <code>kotlinc<\/code>. The draft returned only <code>(0,12)<\/code> for the input above and threw on the empty list. The fixed version returned both sessions and an empty list.<\/p>\n<p>Keep the trace small. Three items exercise every branch here. A ten-item trace burns time and hides the row that matters.<\/p>\n<h2>A sample script for writing and tracing on the board<\/h2>\n<p>The trace only earns credit if the interviewer can follow it. Here&#8217;s what you might say from the clarifying question to the fix. Adapt the names to your problem.<\/p>\n<blockquote>\n<p>Before I write anything, one question. If a session ends at 10 and the next starts at 10, do I merge them? Okay, I&#8217;ll merge. So the plan is to sort by start, then walk once and extend or close the current session. That&#8217;s n log n for the sort.<\/p>\n<p>That&#8217;s my draft. I can&#8217;t run it here, so I&#8217;ll trace it instead. I&#8217;ll use three sessions out of order, two overlapping and one separate. I&#8217;ll keep a table of s, current and result per step.<\/p>\n<p>After step two, result has the merged session from 0 to 12. Current is the session from 20 to 25. Then the loop ends and I return result. So the 20 to 25 session never gets added. That&#8217;s a bug. I&#8217;ll add current to result after the loop.<\/p>\n<p>Now the edge case. An empty list fails at sorted[0], so I&#8217;ll return an empty list first. I&#8217;m fairly sure of sortedBy. If the name is off, it&#8217;s still a sort by start time. Last, I&#8217;d want tests for touching sessions and one session fully inside another.<\/p>\n<\/blockquote>\n<p>Each part of the script does a job. The question up front settles the <code>&lt;=<\/code> before it matters. The plan line gives the interviewer a map, so the code reads as intended. &#8220;I can&#8217;t run it, so I&#8217;ll trace it&#8221; tells the interviewer you&#8217;re verifying on purpose. The table makes every claim checkable. The interviewer can glance at a row and agree or object.<\/p>\n<p>Naming the bug yourself is the strongest moment. You found it, so the interviewer sees testing skill. The hedge on <code>sortedBy<\/code> handles an API you&#8217;re unsure of without stopping. The closing test list shows what you&#8217;d run if you had a machine.<\/p>\n<h2>Ask the recruiter four questions before an in-person coding interview<\/h2>\n<p>Formats vary by company and by round. Companies often don&#8217;t publish the details. So ask before you practice. Google&#8217;s virtual guide says not every interview uses the same tools. It tells candidates to check with the recruiter. Here&#8217;s a short message that covers the gaps.<\/p>\n<blockquote>\n<p>Hi Sam, thanks for confirming the onsite. A few logistics questions so I can prepare well. Which rounds will be in person? Which will be remote? For the coding rounds, will I write on a whiteboard, paper or a laptop? If it&#8217;s a laptop, does it have an editor with autocomplete? And is Kotlin fine for coding, or should I use another language?<\/p>\n<\/blockquote>\n<p>The answers change your practice. A whiteboard means writing large and leaving room for the trace table. Paper means you can&#8217;t erase, so write the plan before the code. A plain laptop editor sits closer to the CodeSignal case. The language answer matters most. Most Android engineers write Kotlin daily. Still, don&#8217;t assume every company accepts it for a general coding round.<\/p>\n<h2>Practice on paper until the trace is a habit<\/h2>\n<p>The trace feels slow the first few times. That&#8217;s the reason to drill it before the onsite. Take problems you&#8217;ve already solved in an IDE, so the algorithm isn&#8217;t the hard part.<\/p>\n<p>For each one, write the solution by hand with a timer running. Then trace two inputs before you check anything. Use one normal input that hits every branch. Add one edge case. Typical edge cases are an empty list, one element or all duplicates. Only then type the code into Android Studio and run it. Count the bugs the run finds that your trace missed. Over a week or two of this, that count should fall.<\/p>\n<p>The same habit helps in the <a href=\"https:\/\/grindloop.ai\/blog\/algorithm-round-android-inter\/\">algorithm round<\/a> when it&#8217;s remote. It also helps in the <a href=\"https:\/\/grindloop.ai\/blog\/ai-assisted-coding-round-android-interviews\/\">AI-assisted coding round<\/a>. There you trace the assistant&#8217;s code instead of your own.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>An in-person coding interview has no run button. How to hand-trace Kotlin on a whiteboard, catch your own bug first, and what to ask the recruiter.<\/p>\n","protected":false},"author":3,"featured_media":848,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"rank_math_title":"In-Person Coding Interview? Your Hand Trace Replaces the Run Button","rank_math_description":"An in-person coding interview has no run button. How to hand-trace Kotlin on a whiteboard, catch your own bug first, and what to ask the recruiter.","rank_math_focus_keyword":"in-person coding interview","footnotes":""},"categories":[20],"tags":[15,44,52,16,12,45],"class_list":["post-779","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-interview-strategy","tag-android","tag-coding-interview","tag-hiring-process","tag-interview-prep","tag-kotlin","tag-technical-interview"],"_links":{"self":[{"href":"https:\/\/grindloop.ai\/blog\/wp-json\/wp\/v2\/posts\/779","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/grindloop.ai\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/grindloop.ai\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/grindloop.ai\/blog\/wp-json\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/grindloop.ai\/blog\/wp-json\/wp\/v2\/comments?post=779"}],"version-history":[{"count":1,"href":"https:\/\/grindloop.ai\/blog\/wp-json\/wp\/v2\/posts\/779\/revisions"}],"predecessor-version":[{"id":780,"href":"https:\/\/grindloop.ai\/blog\/wp-json\/wp\/v2\/posts\/779\/revisions\/780"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/grindloop.ai\/blog\/wp-json\/wp\/v2\/media\/848"}],"wp:attachment":[{"href":"https:\/\/grindloop.ai\/blog\/wp-json\/wp\/v2\/media?parent=779"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/grindloop.ai\/blog\/wp-json\/wp\/v2\/categories?post=779"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/grindloop.ai\/blog\/wp-json\/wp\/v2\/tags?post=779"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}