--- title: Linux kernel now allows AI-generated code, with full... description: AI coding tools have shifted the software development bottleneck from writing code to reviewing it. Real-world data shows AI-assisted PRs merge at less than... canonical: https://daily.dev/posts/linux-kernel-now-allows-ai-generated-code-with-full-responsibility-on-the-human-submitter-vbc4oxiin twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Linux kernel now allows AI-generated code, with full responsibility on the human submitter | daily.dev og:description: AI coding tools have shifted the software development bottleneck from writing code to reviewing it. Real-world data shows AI-assisted PRs merge at less than... og:url: https://daily.dev/posts/linux-kernel-now-allows-ai-generated-code-with-full-responsibility-on-the-human-submitter-vbc4oxiin og:image: https://api.daily.dev/og/posts/vBC4oXiIn.png og:image:alt: Linux kernel now allows AI-generated code, with full responsibility on the human submitter 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. # Linux kernel now allows AI-generated code, with full responsibility on the human submitter **[Collections](https://daily.dev/sources/collections)** · 5 min read · 7 upvotes · 3 comments ## Summary AI coding tools have shifted the software development bottleneck from writing code to reviewing it. Real-world data shows AI-assisted PRs merge at less than half the rate of human-authored ones, code churn has increased, and 45% of AI-generated code contains security flaws. The Linux kernel now requires disclosure of AI assistance via an 'Assisted-by:' tag, with humans bearing full responsibility for submitted AI code. Other projects like NetBSD banned AI code outright, while cURL shut down its bug bounty after AI-generated spam submissions. A key concern is 'comprehension debt' — code that works but no human fully understands — and a reported 19% productivity drop among senior developers. Proposed mitigations include spec-driven development, TypeScript for type safety, shared prompting standards (e.g., versioned AGENTS.md files), and treating AI output like an unreliable dependency requiring validation. ## Content ## The bottleneck has moved For most of software development history, the bottleneck was writing code. AI has shifted that. The new bottleneck is reviewing it. The RavenDB team found this out firsthand. Two self-contained features that would have taken one month and three months respectively were done in two days and one week. That's a real productivity gain. But it came with a catch: someone had to read 30,000 lines of AI-generated code, enforce standards, and evaluate architectural decisions. The writing got faster. The reviewing got harder. LinearB's 2026 Engineering Benchmarks put numbers to this. Despite 88%+ developer AI adoption, AI-assisted pull requests merge at less than half the rate of human-authored code. The code arrives faster. It sits in review longer. AI accelerated generation while exposing every weakness in existing review processes. This is the pattern across the board: AI amplifies what's already there. Strong engineering culture gets stronger. Weak processes get more chaotic, faster. ## What the Linux kernel decided After a six-month debate, the Linux kernel project finalized its policy on AI-generated code. The trigger was a merged patch that turned out to be AI-assisted without disclosure — it caused a performance regression because a `read_mostly` annotation was missing. The kind of subtle, context-specific detail that AI tends to miss. The resulting rules are straightforward: - Contributors must disclose AI assistance using an `Assisted-by:` tag - AI tools cannot add `Signed-off-by` tags — only humans can certify the Developer Certificate of Origin - Human contributors bear full responsibility for any AI-generated code they submit If AI-written code breaks something in the kernel, the blame falls on the human who submitted it. Full stop. Other open source projects landed in different places. NetBSD banned AI-generated code outright. cURL shut down its bug bounty program after being flooded with AI-generated slop submissions. Mesa requires contributors to demonstrate they actually understand any AI-generated code they submit. The Linux approach — allow it, disclose it, own it — sits somewhere in the middle. ## The comprehension problem Here's what doesn't show up in velocity metrics: senior developers are reportedly experiencing a 19% productivity drop as AI-generated code shifts their role toward reviewing code they didn't write and don't fully understand. Call it comprehension debt — code that compiles, passes tests, and looks clean, but where no human actually knows what it's doing or why. The numbers behind this are uncomfortable. Around 45% of AI-generated code contains security flaws. Code churn increased from 5.5% to 7.9%. Duplicated code blocks grew eightfold in 2024. Junior developers using AI score 17% lower on comprehension tests. That last one matters more than it might seem. Junior developers learning by writing code is how senior developers get made. If that pipeline degrades, the people capable of reviewing AI output competently become scarcer over time. The proposed fix is spec-driven development: humans write specifications defining intent and architecture before AI generates anything. The AI fills in the implementation; the human owns the design. TypeScript helps too — it catches around 94% of LLM type errors. Treating AI output like an unreliable dependency, with circuit breakers and fallbacks, is more honest than treating it like a trusted colleague. ## Review as the new quality gate One team went further than most: they changed policy to allow AI tool approval alone to merge pull requests, removing mandatory human code review. Not everyone was comfortable with it. Some team members were excited about the velocity gains; others worried about code quality and whether requirements were actually being met. The norms they landed on: keep all team members notified of PRs, require human review for model and API changes, leave non-trivial logic changes to the author's discretion, and invest more in automated test coverage to compensate for reduced human oversight. Whether that's the right call depends on the team and the codebase. But the underlying tension is real everywhere: human review is the quality gate, and AI is generating more code than review processes were built to handle. The prompting side matters too. AI models fill in missing context with generic patterns that may not fit the specific system. The fix isn't just better review after the fact — it's better prompting upfront, with system context embedded and shared prompting standards across the team. Validation earlier in the workflow is cheaper than debugging later. ## Consistency at scale When every developer uses AI tools differently, output increases but so does variance. Code structure becomes inconsistent. Learning fragments. Performance perceptions get skewed because some people are getting 10x leverage and others are getting 2x, and nobody's sure why. Engineering leaders who treat AI usage conventions like a style guide — a versioned `CLAUDE.md` or `AGENTS.md` with shared baselines — end up scaling their capabilities. Teams without that alignment scale their inconsistencies instead. The developer who comes out ahead in this transition isn't the one who generates the most code. It's the one who owns intent, sets guardrails, and can actually verify correctness — not just that the code runs, but that it does the right thing for the right reasons. ## Community discussion Top comments from developers on daily.dev. **@yildirim** · 3 upvotes > I am trusting the review procedures here so i don't mind if AI slop exists in the kernel code. There were multiple events of human slop pass thru the review so AI or not slop is slop. > > But; > > Human writes, AI reviews - OK (almost) > > AI writes, human reviews - OK > > AI writes, AI reviews - apocalypse happens ## Similar posts on daily.dev - [As AI helps us write more code, who’s catching the bugs?](https://daily.dev/posts/as-ai-helps-us-write-more-code-who-s-catching-the-bugs--swtlsnqbi) · LeadDev · 1 upvotes · 1 comments - [AI-authored code needs more attention, contains worse bugs](https://daily.dev/posts/ai-authored-code-needs-more-attention-contains-worse-bugs-wflwmydbc) · The Register · 0 upvotes · 0 comments --- Tags: [#open-source](https://daily.dev/tags/open-source), [#linux](https://daily.dev/tags/linux) [View this post on daily.dev](https://daily.dev/posts/linux-kernel-now-allows-ai-generated-code-with-full-responsibility-on-the-human-submitter-vbc4oxiin) ```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":"Linux kernel now allows AI-generated code, with full responsibility on the human submitter","url":"https://daily.dev/posts/linux-kernel-now-allows-ai-generated-code-with-full-responsibility-on-the-human-submitter-vbc4oxiin","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/linux-kernel-now-allows-ai-generated-code-with-full-responsibility-on-the-human-submitter-vbc4oxiin"},"datePublished":"2026-04-12T23:18:21.836Z","dateModified":"2026-04-23T23:21:39.767Z","description":"AI coding tools have shifted the software development bottleneck from writing code to reviewing it. Real-world data shows AI-assisted PRs merge at less than...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/ecd75e392b8935d0ed0600f19c4f03c4?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/ecd75e392b8935d0ed0600f19c4f03c4?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Collections","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":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":3,"discussionUrl":"https://daily.dev/posts/linux-kernel-now-allows-ai-generated-code-with-full-responsibility-on-the-human-submitter-vbc4oxiin","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":7},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":3}],"keywords":"open-source,linux","timeRequired":"PT5M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"Linux kernel now allows AI-generated code, with full responsibility on the human submitter"}]} {"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/linux-kernel-now-allows-ai-generated-code-with-full-responsibility-on-the-human-submitter-vbc4oxiin","comment":[{"@type":"Comment","text":"I am trusting the review procedures here so i don’t mind if AI slop exists in the kernel code. There were multiple events of human slop pass thru the review so AI or not slop is slop.\nBut;\nHuman writes, AI reviews - OK (almost)\nAI writes, human reviews - OK\nAI writes, AI reviews - apocalypse happens","datePublished":"2026-04-14T07:31:01.683Z","dateModified":"2026-04-15T13:55:52.525Z","url":"https://daily.dev/posts/vBC4oXiIn#c-mirjeTijX","author":{"@type":"Person","name":"Yasin İsa YILDIRIM","url":"https://daily.dev/yildirim","image":"https://media.daily.dev/image/upload/s--i098KjZN--/f_auto/v1727931829/avatars/avatar_EsTd0u1bqW4Q1cI9yRp5N"},"interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":3}}]} ```