{"slug": "ai-for-cs-assignments-how-i-survive-technical-writing-without-completely-it", "title": "AI for CS Assignments: How I Survive Technical Writing (Without Completely Dreading It)", "summary": "A computer science student describes using AI tools to complete writing assignments for technical courses, arguing it is a pragmatic time allocation decision similar to using Stack Overflow. The student emphasizes the need to provide AI with course-specific context to generate useful drafts, and warns of issues like tone mismatch and missing specificity that require editing.", "body_md": "Sophomore year. My algorithms professor dropped a 1,500-word reflection assignment about the ethics of AI in software development on a Monday morning. The class is called Data Structures and Algorithms. There's not supposed to be a reflection assignment. I stared at the prompt for a full minute, confused about what any of it had to do with binary trees, and then I did what I now do for every writing assignment: I opened an AI tool and started talking to it like it was a very patient tutor who could also write.\n\nI'm a CS student. I'm not bad at school. I'm just bad at writing, and I've stopped pretending that's going to change before I graduate. My brain works in logic, systems, and patterns. Prose is not my thing. Using AI for CS assignments isn't a moral position for me. It's just a time allocation decision, the same way using Stack Overflow to debug is a time allocation decision. I'm not going to build a binary search tree from scratch when documentation exists. I'm not going to write a 1,500-word ethics reflection from scratch when AI exists.\n\nIf you're a CS student who has the same relationship with writing that I do, here's a breakdown of how I approach it.\n\nThe thing nobody tells you before you try this is that AI outputs for technical writing are weirdly inconsistent. I learned this the hard way my freshman year.\n\nFor writing-heavy majors, AI can probably produce a passable essay with minimal editing. For CS writing assignments, that's less reliable. CS professors write prompts that assume you've been paying attention in class. They want you to connect concepts, reference the specific material, and sometimes argue a position using the exact framework they lectured on.\n\nGeneric AI output misses that context completely. I've gotten drafts that were technically correct, coherent, and completely useless because they didn't engage with the actual course content. My professor wanted me to analyze an algorithm design decision using the trade-off framework from week four. The AI gave me a thoughtful general essay about algorithmic complexity. That's not the same thing.\n\nSo the first thing I figured out is that using AI for CS assignments requires you to give it something to work with. A blank prompt gets you a blank-feeling essay. You have to do some upfront work to get useful output.\n\nMy process starts with what I call a context dump. Before I write a single word of the actual prompt, I paste in everything relevant: the assignment instructions, any readings or slides the professor referenced, my own notes from class if they're halfway readable, and a rough bullet list of the points I want to make.\n\nThat last part matters more than it sounds. Even if my bullet points are rough and incomplete, having them there shifts the output from generic to specific. The AI is building on my material instead of inventing its own. The result engages with the course content because I gave it the course content.\n\nFor a 1,500-word assignment, my context dump probably takes 20-30 minutes. That's not nothing, but it's way less than writing from scratch, and the output is significantly better.\n\nSome assignments have less context to work with. Discussion posts, for example, are often just \"respond to this question in 200-300 words.\" Those I treat differently. Less setup, lighter editing. I'm not going to spend 30 minutes preparing for a 200-word discussion post.\n\nGetting a draft is step one. The draft is almost never submittable without editing, and I've gotten into trouble when I treated it like it was.\n\nThe issues I run into consistently are:\n\n**Tone mismatch.** AI writes formally. I write casually. My in-class writing samples exist, and professors aren't stupid. If there's a significant style gap between how I write in a timed quiz and how I write in a take-home essay, someone might notice. So I edit for tone first. I cut the academic vocabulary I'd never use and replace it with how I'd phrase things.\n\n**Missing specificity.** Even with a context dump, AI sometimes produces sentences that are technically accurate but vague. \"This design decision involves trade-offs between time and space complexity\" is not useful. My professor knows that. What they want is for me to say which trade-off, why it matters in this specific context, and what I think about it. I add that layer manually.\n\n**Weird paragraph length.** AI tends to write in uniform paragraph sizes, which looks artificial. I break things up or combine sections based on what makes sense for the argument.\n\n**Transition phrases I'd never say.** \"It is worth noting that...\" No. I delete these and replace them with something that sounds like me or nothing at all.\n\nThe editing pass takes me maybe 30-40 minutes depending on the assignment. So total time investment for a 1,500-word essay is roughly an hour to an hour and a half. That's still significantly less than writing from scratch, which for me would be three or four miserable hours and a worse result.\n\nI want to talk about this because it's the part that adds the most anxiety, and I've learned enough about it to be less anxious than I used to be.\n\nMost universities are using AI detection tools that work by analyzing patterns in text that are statistically associated with AI-generated writing. The tools aren't perfect. They flag human writing sometimes. They miss AI writing sometimes. The threshold matters. A paper that comes back 40% AI-flagged is very different from one that comes back 90% flagged.\n\nMy edited drafts, after the kind of editing I described above, generally come back in a range that doesn't trigger concern. The tone edits and specificity additions change the text enough that the statistical patterns shift. But I still check before I submit anything major, because I'd rather know than not know.\n\nI wrote more about the mechanics of how detection works [in a piece I published a few months ago](https://eliotreads.substack.com/p/i-actually-looked-into-how-ai-detectors). If you want to understand what the tools are doing, that's probably worth reading before you submit something high-stakes.\n\nThe short version is: detection is probabilistic, not certain. A flagged paper is not automatic proof of AI use. An unflagged paper is not a guarantee of safety. The risk is real but it's manageable if you edit seriously.\n\nThere are assignment types where this workflow doesn't work well, and I've learned to identify them early.\n\n**Anything with a specific argument the professor wants.** If the assignment is \"argue for or against position X using the framework from this week's lecture,\" and I haven't internalized that framework, the AI output is going to be hollow. I can tell. My professor will definitely be able to tell.\n\n**Short writing that has to match your in-class voice closely.** Some professors use short reflections throughout the semester partly to build a record of your writing style. Those are the assignments where the gap between an AI-assisted paper and your natural voice is most visible and most risky.\n\n**Anything with citations from sources the professor assigned.** AI will hallucinate citations with confidence. I've caught it inventing papers that don't exist, inventing page numbers for real papers, and attributing quotes to the wrong author. For anything that requires real citations, I pull the sources myself and verify every single one.\n\n**Oral follow-up.** Some professors ask students to come in and discuss their papers. If you can't talk about what you wrote, that's a bad situation. I make sure I understand everything in the draft before I submit it.\n\nThis part I enjoy more than the academic stuff.\n\nMy GitHub READMEs used to be three lines. Project name, what it does, how to run it. That's it. I knew this was bad but I couldn't make myself write more. Now I use the same basic workflow: context dump (the code, what the project does, any interesting technical decisions I made), let the AI draft something, edit for my voice, and publish.\n\nThe result looks like I'm a developer who can explain his own projects. Which is apparently a skill recruiters care about, even though it's not what I thought I was signing up for when I started this degree.\n\nSame with documentation. I'm working on a side project right now that I want to put in my portfolio, and the docs need to be readable. AI handles the first pass. I handle the part that requires knowing what the code does and why I made specific choices.\n\nFor portfolio work, detection is less of a concern. Nobody's running recruiter AI detectors on GitHub READMEs. But sounding human still matters because a robotic README signals low effort on the one artifact that's supposed to show you off.\n\nI think the current cat-and-mouse situation between AI tools and detection tools is going to get more complicated before it gets simpler. Detection is improving. AI output is also improving. Universities are updating their policies semi-regularly, mostly in ways that are inconsistent and hard to keep up with.\n\nThe practical reality for a student right now is that the risk is real but manageable with a serious editing process. The students I've seen get flagged were mostly the ones who ran a bare AI output through submission without editing it. That's not a good strategy.\n\nI also believe the line between \"using AI\" and \"using any other tool to help you work\" is blurry in a way that a lot of institutional AI policies haven't caught up with. I use AI the same way I use syntax highlighters, linters, and autocomplete. It's a tool that makes me faster at a task. The task still requires my judgment, my context, and my editing.\n\nThat's the part that makes AI for CS assignments work when it works. The tool gets you to a draft. The draft is your problem.", "url": "https://wpnews.pro/news/ai-for-cs-assignments-how-i-survive-technical-writing-without-completely-it", "canonical_source": "https://dev.to/eliots/ai-for-cs-assignments-how-i-survive-technical-writing-without-completely-dreading-it-3cjl", "published_at": "2026-07-10 10:44:03+00:00", "updated_at": "2026-07-10 11:13:23.559405+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-tools", "developer-tools"], "entities": ["Stack Overflow"], "alternates": {"html": "https://wpnews.pro/news/ai-for-cs-assignments-how-i-survive-technical-writing-without-completely-it", "markdown": "https://wpnews.pro/news/ai-for-cs-assignments-how-i-survive-technical-writing-without-completely-it.md", "text": "https://wpnews.pro/news/ai-for-cs-assignments-how-i-survive-technical-writing-without-completely-it.txt", "jsonld": "https://wpnews.pro/news/ai-for-cs-assignments-how-i-survive-technical-writing-without-completely-it.jsonld"}}