--- title: 3. How can words turn into actions? | daily.dev description: A dialogue format explains how AI agents turn plain-language requests into structured tool calls. A weather question like 'Will I need a raincoat in Oxford... canonical: https://daily.dev/posts/3-how-can-words-turn-into-actions--xeqanc9zz twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: 3. How can words turn into actions? | daily.dev og:description: A dialogue format explains how AI agents turn plain-language requests into structured tool calls. A weather question like 'Will I need a raincoat in Oxford... og:url: https://daily.dev/posts/3-how-can-words-turn-into-actions--xeqanc9zz og:image: https://api.daily.dev/og/posts/XeQaNc9zZ.png og:image:alt: 3. How can words turn into actions? 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. # 3. How can words turn into actions? **[Planet Python](https://daily.dev/sources/planetpython)** · 17 min read · 0 upvotes · 0 comments ## Summary A dialogue format explains how AI agents turn plain-language requests into structured tool calls. A weather question like 'Will I need a raincoat in Oxford next Saturday?' must become a precise JSON request with a resolved location and date before a tool can use it. The piece walks through failure modes: a request can look valid but target the wrong Oxford, a duration field can be '30' when the user said 'one hour', or a write action can be correctly formed yet lack permission. It stresses that checking required fields alone can't catch these errors, and that better tool-use performance in a model reduces mistakes but doesn't eliminate the need for separate checks. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Why does checking that a tool request has all required fields filled in not guarantee the request is correct? Required-field checks only confirm presence and valid format, not intent match. A request for weather in 'Oxford, Mississippi, US' with a valid date passes every field check yet targets the wrong Oxford if the user meant England. Similarly, a duration of 30 minutes passes a 'must be positive number' check even if the user asked for one hour. Catching these errors requires comparing values against the user's actual instruction, not just validating the tool's input schema. _Developers designing agent validation logic can follow the nuances of tool-call correctness on daily.dev._ ### How did LLMs call external tools before dedicated tool-calling APIs existed? Before dedicated tool-calling interfaces, developers used text conventions, instructing a model to write something like 'WEATHER_REQUEST' followed by the location and date in its plain text reply. Software watched for that convention, parsed it, and executed the corresponding action. Reliability depended entirely on whether the model followed the convention consistently, since there was no structured labeling distinguishing tool requests from ordinary text. _Anyone tracing how tool-calling evolved can follow agent-architecture deep dives on daily.dev._ ### Why can an AI agent change my calendar even if the model itself didn't intend to bypass restrictions? A write action can occur if any available tool or credential path still reaches the calendar service with write access, even after removing a dedicated calendar-writing tool. A properly enforced harness rule rejects write requests regardless of phrasing, but a bypass becomes possible if a general-purpose tool can contact the same service using credentials that permit changes. The strongest protection is a read-only connection, where the service itself refuses writes no matter what the model requests. _Engineers hardening agent permissions can weigh these tradeoffs using resources shared on daily.dev._ ## Similar posts on daily.dev - [2. Who actually does the work, the model or the software?](https://daily.dev/posts/2-who-actually-does-the-work-the-model-or-the-software--xoxmgcac3) · Planet Python · 1 upvotes · 0 comments - [Your Agent Isn't Dumb. Your Tools Are.](https://daily.dev/posts/your-agent-isn-t-dumb-your-tools-are--grledaace) · The T-Shaped Dev · 3 upvotes · 2 comments - [Tool Calling, Explained: How AI Agents Decide What to Do Next](https://daily.dev/posts/tool-calling-explained-how-ai-agents-decide-what-to-do-next-ug95msj09) · Towards Data Science · 1 upvotes · 0 comments - [Your AI Answered. Did Anything Actually Happen?](https://daily.dev/posts/your-ai-answered-did-anything-actually-happen--hjrohzah5) · Medium · 0 upvotes · 0 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#llm](https://daily.dev/tags/llm), [#ai-safety](https://daily.dev/tags/ai-safety) [View this post on daily.dev](https://daily.dev/posts/3-how-can-words-turn-into-actions--xeqanc9zz) ```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":"3. How can words turn into actions?","url":"https://daily.dev/posts/3-how-can-words-turn-into-actions--xeqanc9zz","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/3-how-can-words-turn-into-actions--xeqanc9zz"},"datePublished":"2026-10-29T10:52:41.548Z","dateModified":"2026-10-29T10:53:26.390Z","description":"A dialogue format explains how AI agents turn plain-language requests into structured tool calls. A weather question like 'Will I need a raincoat in Oxford...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/a70e741610f2bee7435adf8545fb58f9?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/a70e741610f2bee7435adf8545fb58f9?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Planet Python","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":"Planet Python","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/eac3cc584cb043b5a242d13752bd3d0c","url":"https://daily.dev/sources/planetpython"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/3-how-can-words-turn-into-actions--xeqanc9zz","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,llm,ai-safety","timeRequired":"PT17M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Planet Python","item":"https://daily.dev/sources/planetpython"},{"@type":"ListItem","position":3,"name":"3. How can words turn into actions?"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/3-how-can-words-turn-into-actions--xeqanc9zz#faq","mainEntity":[{"@type":"Question","name":"Why does checking that a tool request has all required fields filled in not guarantee the request is correct?","acceptedAnswer":{"@type":"Answer","text":"Required-field checks only confirm presence and valid format, not intent match. A request for weather in 'Oxford, Mississippi, US' with a valid date passes every field check yet targets the wrong Oxford if the user meant England. Similarly, a duration of 30 minutes passes a 'must be positive number' check even if the user asked for one hour. Catching these errors requires comparing values against the user's actual instruction, not just validating the tool's input schema. Developers designing agent validation logic can follow the nuances of tool-call correctness on daily.dev."}},{"@type":"Question","name":"How did LLMs call external tools before dedicated tool-calling APIs existed?","acceptedAnswer":{"@type":"Answer","text":"Before dedicated tool-calling interfaces, developers used text conventions, instructing a model to write something like 'WEATHER_REQUEST' followed by the location and date in its plain text reply. Software watched for that convention, parsed it, and executed the corresponding action. Reliability depended entirely on whether the model followed the convention consistently, since there was no structured labeling distinguishing tool requests from ordinary text. Anyone tracing how tool-calling evolved can follow agent-architecture deep dives on daily.dev."}},{"@type":"Question","name":"Why can an AI agent change my calendar even if the model itself didn't intend to bypass restrictions?","acceptedAnswer":{"@type":"Answer","text":"A write action can occur if any available tool or credential path still reaches the calendar service with write access, even after removing a dedicated calendar-writing tool. A properly enforced harness rule rejects write requests regardless of phrasing, but a bypass becomes possible if a general-purpose tool can contact the same service using credentials that permit changes. The strongest protection is a read-only connection, where the service itself refuses writes no matter what the model requests. Engineers hardening agent permissions can weigh these tradeoffs using resources shared on daily.dev."}}]} ```