--- title: The 3 Pillars of the AI Code Paradox (highly technical) description: An analysis argues that AI coding tools won't reduce software spending but increase it, applying Jevons paradox: as the cost per feature drops, companies build... canonical: https://daily.dev/posts/the-3-pillars-of-the-ai-code-paradox-highly-technical--sdqf3m8fi twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: The 3 Pillars of the AI Code Paradox (highly technical) | daily.dev og:description: An analysis argues that AI coding tools won't reduce software spending but increase it, applying Jevons paradox: as the cost per feature drops, companies build... og:url: https://daily.dev/posts/the-3-pillars-of-the-ai-code-paradox-highly-technical--sdqf3m8fi og:image: https://api.daily.dev/og/posts/sDqf3M8Fi.png og:image:alt: The 3 Pillars of the AI Code Paradox (highly technical) 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. # The 3 Pillars of the AI Code Paradox (highly technical) **[Kantan Coding](https://daily.dev/sources/kantancoding)** · 18 min read · 0 upvotes · 0 comments ## Summary An analysis argues that AI coding tools won't reduce software spending but increase it, applying Jevons paradox: as the cost per feature drops, companies build far more features, growing total spend. It then dives into the technical reasons AI-generated code causes complexity problems - explaining tokens, embeddings, and the attention mechanism, then showing how context window limits cause AI to miss cross-module dependencies, introduce hidden performance regressions, and fail on the 'lost in the middle' problem. Sparse attention and retrieval-augmented search index approaches are presented as partial mitigations, each with their own failure modes (e.g., missed strategic token pairs, or code with no explicit reference to a dependent module). The video includes a sponsored segment for Morph, a legacy code modernization AI platform. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### why does AI generated code break things outside the files it was working on AI assistants are limited by a context window - the maximum number of tokens they can read at once - so changes made within that window can silently break modules or dependencies that sit outside it, since the model has no visibility into code it cannot see. A cited example: modifying an order-total calculation broke a nightly revenue report that read from the same database table but wasn't in context, causing discounts to be subtracted twice. _daily.dev surfaces practical writeups on avoiding AI code generation pitfalls like this before they hit production._ ### why doesn't just increasing the AI context window size fix cross-module bugs Larger context windows introduce their own problems: attention computation scales quadratically (O(n^2)) with token count, so doubling the window roughly quadruples compute cost, and models suffer from 'lost in the middle,' a documented weakness where they pay closer attention to the start and end of long text while missing critical details buried in the middle, even when that detail is technically within the window. _developers weighing context window tradeoffs for AI coding tools can track these nuances on daily.dev._ ### what is sparse attention and why doesn't it fully solve AI code comprehension problems Sparse attention has each token attend to only a strategic subset of other tokens instead of comparing every token pair, cutting compute cost versus full attention. The tradeoff is that many token pairs are never compared at all, so if the chosen subset happens to skip an important dependency, a critical connection between modules can still be missed within the context window itself. _daily.dev helps engineers stay current on attention mechanism tradeoffs shaping AI coding tool reliability._ ## Similar posts on daily.dev - [What AI coding costs you](https://daily.dev/posts/what-ai-coding-costs-you-8ii7uzjvf) · Hacker News · 2 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 - [How to solve the AI paradox in software development with intelligent orchestration](https://daily.dev/posts/how-to-solve-the-ai-paradox-in-software-development-with-intelligent-orchestration-qybkwrkho) · The New Stack · 0 upvotes · 0 comments - [AI cuts software developers some slack](https://daily.dev/posts/ai-cuts-software-developers-some-slack-esg9w0zqq) · InfoWorld · 0 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#architecture](https://daily.dev/tags/architecture), [#ai-coding](https://daily.dev/tags/ai-coding), [#rag](https://daily.dev/tags/rag) [View this post on daily.dev](https://daily.dev/posts/the-3-pillars-of-the-ai-code-paradox-highly-technical--sdqf3m8fi) ```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":"The 3 Pillars of the AI Code Paradox (highly technical)","url":"https://daily.dev/posts/the-3-pillars-of-the-ai-code-paradox-highly-technical--sdqf3m8fi","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/the-3-pillars-of-the-ai-code-paradox-highly-technical--sdqf3m8fi"},"datePublished":"2026-09-15T14:29:51.623Z","dateModified":"2026-09-15T14:30:15.294Z","description":"An analysis argues that AI coding tools won't reduce software spending but increase it, applying Jevons paradox: as the cost per feature drops, companies build...","image":"https://i.ytimg.com/vi/k2qls2LiBRc/sddefault.jpg","thumbnailUrl":"https://i.ytimg.com/vi/k2qls2LiBRc/sddefault.jpg","isAccessibleForFree":true,"articleSection":"Kantan Coding","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":"Kantan Coding","logo":"https://media.daily.dev/image/upload/s--FBdOHLfr--/f_auto/v1729363355/logos/kantancoding","url":"https://daily.dev/sources/kantancoding"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/the-3-pillars-of-the-ai-code-paradox-highly-technical--sdqf3m8fi","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,architecture,ai-coding,rag","timeRequired":"PT18M","video":{"@type":"VideoObject","name":"The 3 Pillars of the AI Code Paradox (highly technical)","description":"An analysis argues that AI coding tools won't reduce software spending but increase it, applying Jevons paradox: as the cost per feature drops, companies build...","thumbnailUrl":"https://i.ytimg.com/vi/k2qls2LiBRc/sddefault.jpg","uploadDate":"2026-09-15T14:29:51.623Z","duration":"PT18M","url":"https://api.daily.dev/r/sDqf3M8Fi","embedUrl":"https://www.youtube.com/embed/k2qls2LiBRc"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Kantan Coding","item":"https://daily.dev/sources/kantancoding"},{"@type":"ListItem","position":3,"name":"The 3 Pillars of the AI Code Paradox (highly technical)"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/the-3-pillars-of-the-ai-code-paradox-highly-technical--sdqf3m8fi#faq","mainEntity":[{"@type":"Question","name":"why does AI generated code break things outside the files it was working on","acceptedAnswer":{"@type":"Answer","text":"AI assistants are limited by a context window - the maximum number of tokens they can read at once - so changes made within that window can silently break modules or dependencies that sit outside it, since the model has no visibility into code it cannot see. A cited example: modifying an order-total calculation broke a nightly revenue report that read from the same database table but wasn't in context, causing discounts to be subtracted twice. daily.dev surfaces practical writeups on avoiding AI code generation pitfalls like this before they hit production."}},{"@type":"Question","name":"why doesn't just increasing the AI context window size fix cross-module bugs","acceptedAnswer":{"@type":"Answer","text":"Larger context windows introduce their own problems: attention computation scales quadratically (O(n^2)) with token count, so doubling the window roughly quadruples compute cost, and models suffer from 'lost in the middle,' a documented weakness where they pay closer attention to the start and end of long text while missing critical details buried in the middle, even when that detail is technically within the window. developers weighing context window tradeoffs for AI coding tools can track these nuances on daily.dev."}},{"@type":"Question","name":"what is sparse attention and why doesn't it fully solve AI code comprehension problems","acceptedAnswer":{"@type":"Answer","text":"Sparse attention has each token attend to only a strategic subset of other tokens instead of comparing every token pair, cutting compute cost versus full attention. The tradeoff is that many token pairs are never compared at all, so if the chosen subset happens to skip an important dependency, a critical connection between modules can still be missed within the context window itself. daily.dev helps engineers stay current on attention mechanism tradeoffs shaping AI coding tool reliability."}}]} ```