--- title: Agentic AI doesn’t have a model problem. It has a... description: Agentic AI systems often fail not because of model limitations but because of poor retrieval. On the BrowseComp-Plus benchmark, GPT-4.1 answers 93.49% of... canonical: https://daily.dev/posts/agentic-ai-doesn-t-have-a-model-problem-it-has-a-retrieval-problem--w44jceia1 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Agentic AI doesn’t have a model problem. It has a retrieval problem. | daily.dev og:description: Agentic AI systems often fail not because of model limitations but because of poor retrieval. On the BrowseComp-Plus benchmark, GPT-4.1 answers 93.49% of... og:url: https://daily.dev/posts/agentic-ai-doesn-t-have-a-model-problem-it-has-a-retrieval-problem--w44jceia1 og:image: https://api.daily.dev/og/posts/W44JCEIa1.png og:image:alt: Agentic AI doesn’t have a model problem. It has a retrieval problem. 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. # Agentic AI doesn’t have a model problem. It has a retrieval problem. **[The New Stack](https://daily.dev/sources/newstack)** · 11 min read · 1 upvotes · 0 comments ## Summary Agentic AI systems often fail not because of model limitations but because of poor retrieval. On the BrowseComp-Plus benchmark, GPT-4.1 answers 93.49% of questions correctly when given supporting documents directly, but accuracy collapses to 14.58% when the same model must rely on BM25 search to find evidence itself. Qwen3-32B shows a similar drop, from 83.25% to 3.49%. Agentic workflows amplify this problem because a single task can trigger hundreds or thousands of retrieval calls, making tail latency, capacity, and per-step retrieval accuracy compound across the task. The piece, sponsored by Vespa.ai, argues that retrieval interfaces built for human search behavior don't suit agents, which can use richer query capabilities like filters, structured queries, and multiple rank profiles. It proposes an 'agent-ready schema' and a 'Schema-to-Skill' approach that documents field meanings, rank profiles, and retrieval strategies so agents can construct better queries directly against a system like Vespa, rather than relying on a separate query-translation layer. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How much does retrieval quality affect agentic AI accuracy compared to giving the model the right documents directly? On the BrowseComp-Plus benchmark, GPT-4.1 answers 93.49% of questions correctly when given the labeled supporting documents directly, but accuracy drops to 14.58% when the agent must search a 100,195-document corpus itself using BM25. Qwen3-32B falls from 83.25% to 3.49% under the same two conditions, showing retrieval, not model capability, is often the bottleneck. _daily.dev surfaces practical retrieval and agentic-RAG engineering approaches for developers weighing this trade-off._ ### Why does agentic search put more load on a retrieval system than traditional human search? A single agent task can trigger hundreds or thousands of retrieval operations within a few minutes, since agents decompose tasks into multiple parallel searches and issue follow-up queries as they learn more. This means measured user-request volume can understate actual load, since the retrieval layer processes far more searches than the number of user-facing requests suggests, and tail latency on any one search can slow the whole step. _Teams scaling agent infrastructure can track these retrieval-layer capacity patterns via daily.dev._ ### What is an agent-ready schema and how does Schema-to-Skill help AI agents search more effectively? An agent-ready schema is a description of the fields, controls, and retrieval options an agent can use, including what each field and rank profile actually means, since names like tier or published are ambiguous without context. Schema-to-Skill combines the application's search schema, engine guidance, and application-specific vocabulary into a Markdown skill the agent consults to choose the right filters, fields, and ranking strategy when constructing queries, rather than defaulting to free-text search or relying on a separate query-translation model. _daily.dev helps engineers building agentic RAG pipelines keep up with evolving retrieval-interface design patterns._ ## Similar posts on daily.dev - [Why retrieval quality is becoming the defining challenge in AI agent architecture](https://daily.dev/posts/why-retrieval-quality-is-becoming-the-defining-challenge-in-ai-agent-architecture-jfxpaa9qo) · Vespa Blog · 0 upvotes · 0 comments - [Agentic search models](https://daily.dev/posts/agentic-search-models-qcwmigdv3) · Software Doug · 5 upvotes · 0 comments - [AI retrieval at scale is becoming a systems problem, not a tooling problem](https://daily.dev/posts/ai-retrieval-at-scale-is-becoming-a-systems-problem-not-a-tooling-problem-javxumrcl) · The New Stack · 0 upvotes · 0 comments - [How to Perform Agentic Information Retrieval](https://daily.dev/posts/how-to-perform-agentic-information-retrieval-uij5rwwll) · Towards Data Science · 1 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#ai-agents](https://daily.dev/tags/ai-agents), [#rag](https://daily.dev/tags/rag) [View this post on daily.dev](https://daily.dev/posts/agentic-ai-doesn-t-have-a-model-problem-it-has-a-retrieval-problem--w44jceia1) ```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":"Agentic AI doesn’t have a model problem. It has a retrieval problem.","url":"https://daily.dev/posts/agentic-ai-doesn-t-have-a-model-problem-it-has-a-retrieval-problem--w44jceia1","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/agentic-ai-doesn-t-have-a-model-problem-it-has-a-retrieval-problem--w44jceia1"},"datePublished":"2026-10-15T12:04:51.629Z","dateModified":"2026-10-15T12:05:22.099Z","description":"Agentic AI systems often fail not because of model limitations but because of poor retrieval. On the BrowseComp-Plus benchmark, GPT-4.1 answers 93.49% of...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/d3c78e48c1b58df80b57c8bba1972b09?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/d3c78e48c1b58df80b57c8bba1972b09?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"The New Stack","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":"The New Stack","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/newstack","url":"https://daily.dev/sources/newstack"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/agentic-ai-doesn-t-have-a-model-problem-it-has-a-retrieval-problem--w44jceia1","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,ai-agents,rag","timeRequired":"PT11M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"The New Stack","item":"https://daily.dev/sources/newstack"},{"@type":"ListItem","position":3,"name":"Agentic AI doesn’t have a model problem. It has a retrieval problem."}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/agentic-ai-doesn-t-have-a-model-problem-it-has-a-retrieval-problem--w44jceia1#faq","mainEntity":[{"@type":"Question","name":"How much does retrieval quality affect agentic AI accuracy compared to giving the model the right documents directly?","acceptedAnswer":{"@type":"Answer","text":"On the BrowseComp-Plus benchmark, GPT-4.1 answers 93.49% of questions correctly when given the labeled supporting documents directly, but accuracy drops to 14.58% when the agent must search a 100,195-document corpus itself using BM25. Qwen3-32B falls from 83.25% to 3.49% under the same two conditions, showing retrieval, not model capability, is often the bottleneck. daily.dev surfaces practical retrieval and agentic-RAG engineering approaches for developers weighing this trade-off."}},{"@type":"Question","name":"Why does agentic search put more load on a retrieval system than traditional human search?","acceptedAnswer":{"@type":"Answer","text":"A single agent task can trigger hundreds or thousands of retrieval operations within a few minutes, since agents decompose tasks into multiple parallel searches and issue follow-up queries as they learn more. This means measured user-request volume can understate actual load, since the retrieval layer processes far more searches than the number of user-facing requests suggests, and tail latency on any one search can slow the whole step. Teams scaling agent infrastructure can track these retrieval-layer capacity patterns via daily.dev."}},{"@type":"Question","name":"What is an agent-ready schema and how does Schema-to-Skill help AI agents search more effectively?","acceptedAnswer":{"@type":"Answer","text":"An agent-ready schema is a description of the fields, controls, and retrieval options an agent can use, including what each field and rank profile actually means, since names like tier or published are ambiguous without context. Schema-to-Skill combines the application's search schema, engine guidance, and application-specific vocabulary into a Markdown skill the agent consults to choose the right filters, fields, and ranking strategy when constructing queries, rather than defaulting to free-text search or relying on a separate query-translation model. daily.dev helps engineers building agentic RAG pipelines keep up with evolving retrieval-interface design patterns."}}]} ```