--- title: Vetted AI code is hard to justify | daily.dev description: A game developer reflects on using a frontier AI coding agent to build a large performance optimization. Planning took days, but comprehending and reviewing... canonical: https://daily.dev/posts/vetted-ai-code-is-hard-to-justify-uoloz8gr7 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Vetted AI code is hard to justify | daily.dev og:description: A game developer reflects on using a frontier AI coding agent to build a large performance optimization. Planning took days, but comprehending and reviewing... og:url: https://daily.dev/posts/vetted-ai-code-is-hard-to-justify-uoloz8gr7 og:image: https://api.daily.dev/og/posts/UoLOz8GR7.png og:image:alt: Vetted AI code is hard to justify og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Vetted AI code is hard to justify **[Lobsters](https://daily.dev/sources/lobsters)** · 1 min read · 0 upvotes · 0 comments ## Summary A game developer reflects on using a frontier AI coding agent to build a large performance optimization. Planning took days, but comprehending and reviewing the massive diff line-by-line took weeks and led to burnout, since the developer insisted on fully understanding every part of the generated code. The author contrasts this with building it unassisted, which would have taken longer but spread the comprehension effort out over time, causing less burnout. The piece ends with a self-made interactive visualization illustrating the relationship between coding speed, comprehension cost, and burnout when using AI-generated code. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Community take How the wider developer community reacted, aggregated from 2 discussions and 19 comments across lobsters, hackernews (as of 2026-08-19). **TL;DR:** Discussion largely agrees the pain described stems from letting an agent produce one giant diff instead of breaking work into small, reviewable, TDD-driven increments; some also found the closing interactive visualization ironic given the essay's anti-AI-vibes stance. **Sentiment:** 15% positive · 55% mixed · 30% skeptical **The case for** - Using structured skill scaffolding (e.g. TDD, planning docs, mermaid diagrams) with agents reportedly produces much better structured, reviewable commits. - One commenter reports genuinely positive results applying best practices with an LLM. - One commenter argues it's still twice as fast as the manual estimate and possibly better tested. **The pushback** - Several argue the root problem was letting the agent produce one massive diff instead of small stacked PRs reviewed incrementally. - Some find it ironic/hypocritical that the author complains AI vetting is exhausting yet used AI to casually generate the closing visualization. - One commenter notes reviewing large amounts of smart-but-flawed AI code carries a higher burnout multiplier than reviewing a human teammate's work because all ownership falls on the reviewer. - One commenter felt the interactive dashboard itself was confusing and would work better as a plain essay with explained quadrants. - Frontier labs are criticized for not baking good engineering practices (small diffs, TDD, DRY) into their harnesses/model training by default. **By community** - lobsters (mixed): Most argue the burnout was avoidable via smaller stacked diffs and skill scaffolding, while others sympathize with the author and note some found the visualization ironic given the essay's premise. - hackernews (mixed): No comments were provided, so no discernible discussion sentiment. **Hottest debate:** Whether the reviewing burnout is fundamentally avoidable by breaking work into smaller, incrementally reviewed diffs, or whether it's an irreducible cost of vetting large amounts of AI-generated code regardless of chunking. **Open questions** - Would breaking the work into smaller stacked diffs and using skill scaffolding have actually prevented the burnout in this specific case? - Is the exhausting part fundamentally about diff size, or about the psychological burden of sole ownership over reviewing AI-authored code versus human teammate code? **Highlights** > > about a week to comprehend the massive diff it produced It is insane to me that models aren't better trained on this and the harnesses don't default to better practices. But you **do not need to have one massive diff**. There are skill plugins [like superpowers](https://github.com/obra/superpowers) that attempt to let these models/harnesses make use of good practices like TDD, DRY, etc and they actually do improve the quality output in a remarkable way and give you much better structured commits. Most people I know who use AI *somewhat* successfully in their workflow have had to spend A LOT of time putting up a ton of skill scaffolding and other instructions before they got to that point. Some of that comes down to project specific idiosyncrasies but a majority from what I have seen is the same sort of stuff as the superpowers skill set, trying to get these things to follow basic fundamental software development practices. As I said, it is borderline insane to me that *none* of the frontier labs includes those practices themselves in their harnesses or model training to any significant degree as far as I can tell. They all seem to focus on the sort of quick idea to code type development that works for prototypes but falls apart for larger projects. And to be extra clear, even with these skills in place it still is exhausting to me to use agentic coding with harnesess like claude code. But if your work requires you to use them or you want to actually trial them, do yourself a favor and at least try and use something like superpowers to get some semblance of sanity in the flow. Now to get back to the article. I find the contrast between *"using AI is extremely exhausting"* and *"Also, here is an interactive representation I had the AI make"* pretty hilarious to be honest. > — [creesch on lobsters · 9 points, 4 comments](https://lobste.rs/s/os20fa/vetted_ai_code_is_hard_justify#c_7b244j) > The problem that I am seeing is less that the feature cannot be broken into smaller pieces, and more that reviewing a nontrivial amount of smart-but-sometimes-very-flawed code has a high burnout multiplier. It feels genuinely different than a human teammate because all of the ownership falls on me, versus being shared with the teammate. > — [amoffat on lobsters · 1 points, 1 comments](https://lobste.rs/s/os20fa/vetted_ai_code_is_hard_justify#c_ns9ypy) > This is a nice way to partition LLM-assisted work into segments based on practical and psychological toll but I find the "dashboard" to be confusing. It's hard to understand all of the sliders without mousing over the tooltips and it's not clear what's happening when you drag one of the sliders and (e.g.) the y-axis scale shifts. I believe the piece would be stronger as a longer essay without the interactive bit. Perhaps you could write brief paragraphs explaining the different ~quadrants~ regimes (I don't know the actual number) that the reader now has to search for by manually searching the 4-dimensional parameter space. > — [n1000 on lobsters · 4 points](https://lobste.rs/s/os20fa/vetted_ai_code_is_hard_justify#c_ardfao) > > Now to get back to the article. I find the contrast between "using AI is extremely exhausting" and "Also, here is an interactive representation I had the AI make" pretty hilarious to be honest. To me it reads like a contrast between having LLMs implement something to the author's production standard — and having an LLM one-shot vibecode a sketch, with no care for precision because all numbers are purely vibes from the beginning. > — [k749gtnc9l3w on lobsters · 2 points](https://lobste.rs/s/os20fa/vetted_ai_code_is_hard_justify#c_duwb6t) > > about a week to comprehend the massive diff it produced I mean here's your problem. Instead of having the LLM crap out one massive change, spend some time breaking the task up. Then have the LLM produce small focused deliverables as pull requests, and review them individually. This is no different from how you'd work on a feature on a team as well I might add. Nobody wants to see massive PRs that conflate a bunch of features together. There's absolutely no reason to let the agent churn out a bunch of code before you look at it. > — [Yogthos on lobsters · 2 points, 2 comments](https://lobste.rs/s/os20fa/vetted_ai_code_is_hard_justify#c_lblktf) **Source threads** - [lobsters](https://lobste.rs/s/os20fa/vetted_ai_code_is_hard_justify) · 20 points · 19 comments - [hackernews](https://news.ycombinator.com/item?id=49329292) · 2 points · 0 comments ## Similar posts on daily.dev - [A different kind of tired](https://daily.dev/posts/a-different-kind-of-tired-cxnlxorqm) · Ruby Flow · 0 upvotes · 0 comments - [Managing the Hidden Overhead of AI Software Engineering](https://daily.dev/posts/managing-the-hidden-overhead-of-ai-software-engineering-odnh0uxvl) · DEV · 1 upvotes · 0 comments - [After two years of vibecoding, I'm back to writing by hand](https://daily.dev/posts/after-two-years-of-vibecoding-i-m-back-to-writing-by-hand-mvxk3c8r1) · Hacker News · 82 upvotes · 14 comments --- Tags: [#game-development](https://daily.dev/tags/game-development), [#code-review](https://daily.dev/tags/code-review) [View this post on daily.dev](https://daily.dev/posts/vetted-ai-code-is-hard-to-justify-uoloz8gr7) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Vetted AI code is hard to justify","url":"https://daily.dev/posts/vetted-ai-code-is-hard-to-justify-uoloz8gr7","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/vetted-ai-code-is-hard-to-justify-uoloz8gr7"},"datePublished":"2026-08-17T22:22:13.722Z","dateModified":"2026-08-19T12:02:46.090Z","description":"A game developer reflects on using a frontier AI coding agent to build a large performance optimization. Planning took days, but comprehending and reviewing...","image":"https://media.daily.dev/image/upload/s--VDukGCjf--/f_auto/v1722860399/public/Placeholder%2002","thumbnailUrl":"https://media.daily.dev/image/upload/s--VDukGCjf--/f_auto/v1722860399/public/Placeholder%2002","isAccessibleForFree":true,"articleSection":"Lobsters","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Lobsters","logo":"https://media.daily.dev/image/upload/s--tl8v_Fku--/f_auto,t_logo/v1698841318/logos/lobste.jpg","url":"https://daily.dev/sources/lobsters"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/vetted-ai-code-is-hard-to-justify-uoloz8gr7","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"game-development,code-review","timeRequired":"PT1M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Lobsters","item":"https://daily.dev/sources/lobsters"},{"@type":"ListItem","position":3,"name":"Vetted AI code is hard to justify"}]} ```