--- title: Only believe what you can validate: a verification... description: A verification framework for agentic AI output is proposed, centered on the idea that AI-generated results should only be trusted when they can be validated.... canonical: https://daily.dev/posts/only-believe-what-you-can-validate-a-verification-framework-for-agentic-ai-ojzgwtdhs twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Only believe what you can validate: a verification framework for agentic AI | daily.dev og:description: A verification framework for agentic AI output is proposed, centered on the idea that AI-generated results should only be trusted when they can be validated.... og:url: https://daily.dev/posts/only-believe-what-you-can-validate-a-verification-framework-for-agentic-ai-ojzgwtdhs og:image: https://api.daily.dev/og/posts/OjZgwTdHS.png og:image:alt: Only believe what you can validate: a verification framework for agentic AI 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. # Only believe what you can validate: a verification framework for agentic AI **[DevBlogs](https://daily.dev/sources/devblogs)** · 11 min read · 0 upvotes · 0 comments ## Summary A verification framework for agentic AI output is proposed, centered on the idea that AI-generated results should only be trusted when they can be validated. The piece describes three failure modes (misses, hallucinations, misinterpretations), the limits of context windows causing 'context rot', and the failure patterns of human-AI-deterministic tool combinations. It presents a six-question iterative framework covering what to check, cost, who performs it, what it proves, the resulting artifact, and available tools, then applies it to two concrete domains: reverse engineering of legacy COBOL systems and AI-assisted code generation, with detailed verification layer tables for each including specific tools like SonarQube, Snyk, JMeter, and Playwright. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Why can't I just trust an AI agent's reverse-engineered documentation of a legacy COBOL module because it looks good on first glance? Because what is easy to verify at a glance (like the first 10-20 lines of a record layout) is also easy for the model to get right, so a quick scan tells you nothing about the accuracy of the rest of a long document. A 2500-word reverse engineering document generated in five minutes can contain hallucinations, misinterpretations, or missing details that only surface after hours of thorough review. _Teams weighing AI-assisted modernization risk track verification practices like this on daily.dev._ ### What causes an AI model to produce confident but wrong output as its context window fills up? Context rot: below 50% of context capacity, middle tokens degrade so the model retains the beginning and end of the input but loses what's in between; above 50% capacity, even the beginning starts to decay, yet the model keeps producing output with the same confident tone throughout, making the degradation invisible to the reader. _Developers debugging inconsistent long-context AI outputs follow explanations like this on daily.dev._ ### What are the three main failure modes when verifying AI-generated output in application modernization? The miss (the AI overlooked something that should be included, leaving output incomplete though nothing is technically wrong), the hallucination (the AI produced something plausible that doesn't exist or isn't true), and the misinterpretation (the AI followed vague instructions to the best of its knowledge but still produced the wrong outcome). _Anyone designing QA processes for agentic AI can compare failure taxonomies like this on daily.dev._ --- Tags: [#testing](https://daily.dev/tags/testing), [#agentic-ai](https://daily.dev/tags/agentic-ai) [View this post on daily.dev](https://daily.dev/posts/only-believe-what-you-can-validate-a-verification-framework-for-agentic-ai-ojzgwtdhs) ```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":"Only believe what you can validate: a verification framework for agentic AI","url":"https://daily.dev/posts/only-believe-what-you-can-validate-a-verification-framework-for-agentic-ai-ojzgwtdhs","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/only-believe-what-you-can-validate-a-verification-framework-for-agentic-ai-ojzgwtdhs"},"datePublished":"2026-09-02T13:54:19.894Z","dateModified":"2026-09-02T13:58:11.320Z","description":"A verification framework for agentic AI output is proposed, centered on the idea that AI-generated results should only be trusted when they can be validated....","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/f0bf968b05394f8be287756ece69c212?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/f0bf968b05394f8be287756ece69c212?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"DevBlogs","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":"DevBlogs","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/8f68b453325f482ebeb73fb780092713","url":"https://daily.dev/sources/devblogs"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/only-believe-what-you-can-validate-a-verification-framework-for-agentic-ai-ojzgwtdhs","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"testing,agentic-ai","timeRequired":"PT11M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"DevBlogs","item":"https://daily.dev/sources/devblogs"},{"@type":"ListItem","position":3,"name":"Only believe what you can validate: a verification framework for agentic AI"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/only-believe-what-you-can-validate-a-verification-framework-for-agentic-ai-ojzgwtdhs#faq","mainEntity":[{"@type":"Question","name":"Why can't I just trust an AI agent's reverse-engineered documentation of a legacy COBOL module because it looks good on first glance?","acceptedAnswer":{"@type":"Answer","text":"Because what is easy to verify at a glance (like the first 10-20 lines of a record layout) is also easy for the model to get right, so a quick scan tells you nothing about the accuracy of the rest of a long document. A 2500-word reverse engineering document generated in five minutes can contain hallucinations, misinterpretations, or missing details that only surface after hours of thorough review. Teams weighing AI-assisted modernization risk track verification practices like this on daily.dev."}},{"@type":"Question","name":"What causes an AI model to produce confident but wrong output as its context window fills up?","acceptedAnswer":{"@type":"Answer","text":"Context rot: below 50% of context capacity, middle tokens degrade so the model retains the beginning and end of the input but loses what's in between; above 50% capacity, even the beginning starts to decay, yet the model keeps producing output with the same confident tone throughout, making the degradation invisible to the reader. Developers debugging inconsistent long-context AI outputs follow explanations like this on daily.dev."}},{"@type":"Question","name":"What are the three main failure modes when verifying AI-generated output in application modernization?","acceptedAnswer":{"@type":"Answer","text":"The miss (the AI overlooked something that should be included, leaving output incomplete though nothing is technically wrong), the hallucination (the AI produced something plausible that doesn't exist or isn't true), and the misinterpretation (the AI followed vague instructions to the best of its knowledge but still produced the wrong outcome). Anyone designing QA processes for agentic AI can compare failure taxonomies like this on daily.dev."}}]} ```