{"id":978,"date":"2026-11-10T09:55:00","date_gmt":"2026-11-10T08:55:00","guid":{"rendered":"https:\/\/grindloop.ai\/blog\/?p=978"},"modified":"2026-11-09T16:33:40","modified_gmt":"2026-11-09T15:33:40","slug":"take-home-follow-up-interview","status":"publish","type":"post","link":"https:\/\/grindloop.ai\/blog\/take-home-follow-up-interview\/","title":{"rendered":"Take-Home Follow-Up Interview: Map the Changes You&#8217;ll Be Asked to Make"},"content":{"rendered":"<p>A take-home follow-up interview is where the code you submitted gets scored. The reviewer has already read your repo. In the call, they ask you to walk through it and defend your choices. Then they ask you to change it, live. Narrative&#8217;s published hiring process says its interviewers propose hypothetical changes to the problem. Jacob Kaplan-Moss&#8217;s guide to take-home exercises suggests asking how you&#8217;d solve it without a library you used. Most of your prep should go into that change. Before the call, list the three changes a reviewer is most likely to ask for. For each one, write down the files it touches and what could break. Then ask your recruiter whether AI tools are allowed in the live session. The take-home rules may not carry over. Below, one Android take-home goes through a 45-minute follow-up. You get a full change map and a script for the live edit.<\/p>\n<h2>Most follow-up questions ask you to change your code<\/h2>\n<p>Start with what companies say they do in this session. <a href=\"https:\/\/jobs.narrative.io\/process\/dev-hiring-process\" target=\"_blank\" rel=\"noopener\">Narrative&#8217;s dev hiring process<\/a> was last updated in February 2026. It has three stages. A phone screen comes first, then a take-home, then a follow-up with the team. You only reach the follow-up if the take-home meets basic requirements. In it, interviewers ask about the operational, scaling and maintainability consequences of your design. They ask what you&#8217;d change before shipping it to production. They also propose hypothetical changes to the problem. Even the written review checks one thing in advance. Did you call out the issues you skipped to save time?<\/p>\n<p>Jacob Kaplan-Moss covered the interviewer&#8217;s side in <a href=\"https:\/\/jacobian.org\/2021\/nov\/23\/wst-homework\/\" target=\"_blank\" rel=\"noopener\">a November 2021 post on coding homework<\/a>. Someone on the team who read your code runs it. He suggests 30 to 45 minutes. It opens with &#8220;walk me through&#8221; your code. His sample follow-ups then turn to change. What would you do differently next time? How would you solve it without a third-party module? Would it still work with a far larger data file?<\/p>\n<p>In both accounts, explaining what you built is the warm-up. Most probes after it are changes or counterfactuals. Each one checks whether you can still steer the code you sent.<\/p>\n<h2>Code you can&#8217;t change on the spot is the trap, whoever wrote it<\/h2>\n<p>Take the GitHub repository browser brief from our post on <a href=\"https:\/\/grindloop.ai\/blog\/what-the-android-take-home-assignment-actually-tests\/\">what the Android take-home assignment tests<\/a>. You fetch a user&#8217;s repositories, show them in a list and open a detail screen on tap. Your submission has one module and one ViewModel per screen. It uses Compose, Hilt and a Room cache for offline use. You sent it on a Friday. The follow-up lands ten days later. By then you&#8217;ve forgotten how the refresh writes to Room.<\/p>\n<p>Two kinds of code fail in this session. The first is code you over-built. Every extra layer is another place a live change has to pass through. The second is code you accepted without fully reading it. That includes generated code. It reads fine in a repo. It falls apart when someone points at line 40 and asks what happens if it changes.<\/p>\n<p>AI rules make this sharper. <a href=\"https:\/\/www.anthropic.com\/candidate-ai-guidance\" target=\"_blank\" rel=\"noopener\">Anthropic&#8217;s candidate guidance<\/a> was last updated in July 2025. It asks candidates to complete take-homes without Claude unless the company says otherwise. Live interviews are &#8220;all you,&#8221; with no AI help unless the company says otherwise. Rules differ by company. But even where a take-home allowed AI, the live session may be you and an editor. Then code you didn&#8217;t steer at home costs you in the follow-up.<\/p>\n<h2>A change map turns the likely probes into rehearsed edits<\/h2>\n<p>A change map is one page you write the day before the call. Reread your own code first. Then predict three changes. Take one from your README&#8217;s &#8220;what I&#8217;d do next&#8221; list. Take one from scale, because both sources above ask about it. Take one from a dependency, because Kaplan-Moss asks about removing one. For each change, write four things. Name the files it touches. Name what could break. Write your plan. Then add the one thing to say before you type.<\/p>\n<p>Here&#8217;s the map for the repository browser.<\/p>\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"\">Change 1: Favorite a repository. Favorites survive offline and restarts.\n  Touches:    a new FavoriteEntity table, RepoDao, RepoRepository,\n              DetailViewModel, DetailScreen.\n  Could break: refresh replaces whole RepoEntity rows. A flag stored on\n              that row would reset on every refresh.\n  Plan:       store favorites in their own table, keyed by repository ID.\n              Bump the database version. Use a destructive fallback,\n              because it&#039;s a cache, and say so.\n  Test first: a repository test that a refresh keeps a favorite.\n\nChange 2: A user with 5,000 repositories.\n  Touches:    the API call, RepoDao, ListViewModel, ListScreen.\n  Could break: today only the first page of results loads.\n  Plan:       Paging 3 with a RemoteMediator, so Room stays the source.\n  Say first:  &quot;Right now this shows 30 repositories and stops.&quot;\n\nChange 3: Remove Hilt.\n  Touches:    the Application class and two ViewModel factories.\n  Plan:       one AppContainer that builds the repository once.\n  Say first:  &quot;One module, so this is small.&quot;<\/pre>\n<p>Each plan follows Android&#8217;s own guidance. Adding a table changes the schema. Room&#8217;s <a href=\"https:\/\/developer.android.com\/training\/data-storage\/room\/migrating-db-versions\" target=\"_blank\" rel=\"noopener\">migration guide<\/a> covers the version bump. It also covers the destructive fallback. That option wipes the data instead of migrating it. That trade is fine for a cache and wrong for user data, so name it out loud. The 30 comes from the GitHub API. Its <a href=\"https:\/\/docs.github.com\/en\/rest\/repos\/repos#list-repositories-for-a-user\" target=\"_blank\" rel=\"noopener\">list repositories endpoint<\/a> returns 30 per page by default. Paging 3 has a <a href=\"https:\/\/developer.android.com\/topic\/libraries\/architecture\/paging\/v3-network-db\" target=\"_blank\" rel=\"noopener\">network and database guide<\/a> built around RemoteMediator. The AppContainer comes from the <a href=\"https:\/\/developer.android.com\/training\/dependency-injection\/manual\" target=\"_blank\" rel=\"noopener\">manual dependency injection guide<\/a>.<\/p>\n<p>Write the &#8220;could break&#8221; line most carefully. It&#8217;s the part of the change you&#8217;d otherwise find live, with someone watching. Change 1 is the clearest case. A favorite flag on the repository row looks like the obvious edit. It would also be silently wiped by the next refresh.<\/p>\n<h2>A sample script for a 45-minute take-home follow-up interview<\/h2>\n<p>Here&#8217;s how the map plays out in the call. Adapt the details to your own submission.<\/p>\n<p>Open with a short tour that hands over control.<\/p>\n<blockquote>\n<p>&#8220;I&#8217;ll give a quick tour, then you steer. It&#8217;s one module. There&#8217;s a repository class, one ViewModel per screen and Compose UI. The repository serves Room first, then refreshes from the network. Three calls are worth pushing on. I used Room for the offline cache. I used Hilt over hand-written wiring. And I skipped a domain layer. I also know two gaps. Only the first page of results loads. There are no UI tests. Where would you like to start?&#8221;<\/p>\n<\/blockquote>\n<p>Say the interviewer asks for favorites that survive a restart. Name the risk before you type.<\/p>\n<blockquote>\n<p>&#8220;This touches four places. One risk comes first. My refresh replaces whole rows, so a flag on the row would reset. I&#8217;ll keep favorites in their own table, keyed by repository ID. That leaves the refresh alone. The schema changes, so the database version goes up. It&#8217;s a cache, so I&#8217;ll use a destructive fallback here. In production I&#8217;d write a migration. I&#8217;ll start with a test that a refresh keeps a favorite. Then I&#8217;ll wire the button. If we run short, I&#8217;ll stop after the test and talk through the UI.&#8221;<\/p>\n<\/blockquote>\n<p>Then the scale question arrives. Give the current limit first, then the plan.<\/p>\n<blockquote>\n<p>&#8220;Right now it shows 30 repositories and stops. That&#8217;s the API&#8217;s default page. For 5,000, I&#8217;d move the list to Paging 3 with a RemoteMediator. Room stays the source of truth. The DAO would return a PagingSource. The ViewModel would expose a paging flow. I wouldn&#8217;t build that in this call. I can sketch where each piece goes.&#8221;<\/p>\n<\/blockquote>\n<h2>Why each part of the script works<\/h2>\n<p>The tour ends with a question, so the interviewer picks the ground. It also names the gaps before they&#8217;re found. Narrative&#8217;s review looks for exactly that. A gap you named reads as a choice. A gap they find reads as a miss.<\/p>\n<p>The favorites answer names the risk before any code. That proves you know how your own pieces connect. That is what a live change tests. Starting with a test also gives you a clean place to stop. If time runs out, you&#8217;ve still shipped the part that proves the hard case.<\/p>\n<p>The scale answer starts with an honest limit and a number. Then it places each new piece in a file you already have. Kaplan-Moss&#8217;s larger-data question is after the same thing. Can you see where your design stops working? Showing where the fix would go is enough.<\/p>\n<h2>Two questions to send your recruiter before the follow-up<\/h2>\n<p>Ask how the session runs and what tools you get. A follow-up can be discussion only. It can also mean editing on your own machine with a shared screen. If you&#8217;ll code live, ask whether you can use your own editor. Our post on <a href=\"https:\/\/grindloop.ai\/blog\/android-studio-coding-interview\/\">using Android Studio in a coding interview<\/a> covers why the answer varies. Then ask about AI tools in the live session, even if the take-home allowed them.<\/p>\n<blockquote>\n<p>&#8220;Thanks for setting up the follow-up. Two quick questions so I can prepare. Will I edit my submission live? If so, can I use my own Android Studio with screen share? And are AI assistants allowed during the session, or only for the take-home?&#8221;<\/p>\n<\/blockquote>\n<p>Send it as soon as the follow-up is booked. Then write your change map the day before. Some companies do allow AI live. Our post on the <a href=\"https:\/\/grindloop.ai\/blog\/ai-assisted-coding-round-android-interviews\/\">AI-assisted coding round<\/a> covers what that changes.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>A take-home follow-up interview asks you to change your own code live. Build a change map first, check the AI rules and use this Android sample script.<\/p>\n","protected":false},"author":3,"featured_media":980,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"rank_math_title":"Take-Home Follow-Up Interview: Map the Changes First","rank_math_description":"A take-home follow-up interview asks you to change your own code live. Build a change map first, check the AI rules and use this Android sample script.","rank_math_focus_keyword":"take-home follow-up interview","footnotes":""},"categories":[20],"tags":[15,52,16,33,45],"class_list":["post-978","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-interview-strategy","tag-android","tag-hiring-process","tag-interview-prep","tag-take-home-assignment","tag-technical-interview"],"_links":{"self":[{"href":"https:\/\/grindloop.ai\/blog\/wp-json\/wp\/v2\/posts\/978","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=978"}],"version-history":[{"count":1,"href":"https:\/\/grindloop.ai\/blog\/wp-json\/wp\/v2\/posts\/978\/revisions"}],"predecessor-version":[{"id":979,"href":"https:\/\/grindloop.ai\/blog\/wp-json\/wp\/v2\/posts\/978\/revisions\/979"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/grindloop.ai\/blog\/wp-json\/wp\/v2\/media\/980"}],"wp:attachment":[{"href":"https:\/\/grindloop.ai\/blog\/wp-json\/wp\/v2\/media?parent=978"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/grindloop.ai\/blog\/wp-json\/wp\/v2\/categories?post=978"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/grindloop.ai\/blog\/wp-json\/wp\/v2\/tags?post=978"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}