{"id":2040,"date":"2025-04-29T04:38:16","date_gmt":"2025-04-29T04:38:16","guid":{"rendered":"https:\/\/elink.cat\/blog\/?p=2040"},"modified":"2026-10-07T21:21:59","modified_gmt":"2026-10-07T21:21:59","slug":"agents-classics-vs-agents-basats-en-llm-dues-maneres-de-pensar","status":"publish","type":"post","link":"https:\/\/elink.cat\/blog\/agents-classics-vs-agents-basats-en-llm-dues-maneres-de-pensar\/","title":{"rendered":"Agents cl\u00e0ssics vs. agents basats en LLM: dues maneres de pensar"},"content":{"rendered":"<span class=\"span-reading-time rt-reading-time\" style=\"display: block;\"><span class=\"rt-label rt-prefix\">Temps de lectura: <\/span> <span class=\"rt-time\"> 2<\/span> <span class=\"rt-label rt-postfix\">minuts<\/span><\/span><p class=\"p5\">Com tots ja sabeu, el m\u00f3n dels <span class=\"s3\"><b>agents d\u2019intel\u00b7lig\u00e8ncia artificial<\/b><\/span> ha viscut una revoluci\u00f3 amb l\u2019aparici\u00f3 dels <span class=\"s3\"><b>LLMs (Large Language Models)<\/b><\/span>. Per\u00f2 abans que aquests models entressin en escena, ja existien agents d\u2019IA que funcionaven amb regles i sistemes m\u00e9s tradicionals.<\/p>\r\n<p class=\"p5\">Contrastem aquests <span class=\"s3\"><b>agents cl\u00e0ssics<\/b><\/span> amb els <span class=\"s3\"><b>agents basats en LLM<\/b><\/span>, per entendre com pensen, com actuen i per qu\u00e8 aquests nous agents estan transformant la forma com entenem la IA.<\/p>\r\n\r\n<h4 class=\"p6\"><b>Qu\u00e8 \u00e9s un agent cl\u00e0ssic?<\/b><b><\/b><\/h4>\r\n<p class=\"p5\">Els <span class=\"s3\"><b>agents cl\u00e0ssics<\/b><\/span> s\u00f3n sistemes dissenyats per operar en entorns definits mitjan\u00e7ant <span class=\"s3\"><b>regles, l\u00f2gica i planificaci\u00f3 programada<\/b><\/span>.<\/p>\r\n<p class=\"p7\"><b>Caracter\u00edstiques:<\/b><b><\/b><\/p>\r\n\r\n<ul>\r\n \t<li class=\"p8\">Segueixen <span class=\"s3\"><b>regles fixes<\/b><\/span> (IF\u2026 THEN\u2026)<\/li>\r\n \t<li class=\"p8\">Utilitzen planificaci\u00f3 formal (com A* o algoritmes heur\u00edstics)<\/li>\r\n \t<li class=\"p9\"><span class=\"s1\">S\u00f3n <\/span><b>predictibles<\/b><span class=\"s1\"> i <\/span><b>deterministes<\/b><b><\/b><\/li>\r\n \t<li class=\"p8\">Funcionen b\u00e9 en entorns <span class=\"s3\"><b>acotats<\/b><\/span> i amb dades estructurades<\/li>\r\n<\/ul>\r\n<p class=\"p5\"><i>Exemple:<\/i> Un robot que navega per un magatzem seguint una ruta predefinida optimitzada per evitar obstacles.<\/p>\r\n\r\n<h4 class=\"p6\"><b>Qu\u00e8 \u00e9s un agent basat en LLM?<\/b><b><\/b><\/h4>\r\n<p class=\"p5\">Els agents <span class=\"s3\"><b>basats en LLMs<\/b><\/span> (com GPT-4, Claude o DeepSeek) poden <span class=\"s3\"><b>entendre llenguatge natural<\/b><\/span>, <span class=\"s3\"><b>raonar<\/b><\/span> i <span class=\"s3\"><b>prendre decisions<\/b><\/span> a partir de contextos no estructurats. Aquests agents actuen com una mena de \u201ccervell flexible\u201d que pot adaptar-se a una gran varietat de situacions.<\/p>\r\n<p class=\"p7\"><b>Caracter\u00edstiques:<\/b><b><\/b><\/p>\r\n\r\n<ul>\r\n \t<li class=\"p8\">Entenen i generen <span class=\"s3\"><b>llenguatge natural<\/b><b><\/b><\/span><\/li>\r\n \t<li class=\"p8\">Aprenen de grans volums de dades<\/li>\r\n \t<li class=\"p8\">S\u00f3n <span class=\"s3\"><b>probabil\u00edstics<\/b><\/span> i adaptatius<\/li>\r\n \t<li class=\"p8\">Poden <span class=\"s3\"><b>raonar en temps real<\/b><\/span>, fer preguntes, planificar i auto-reflexionar<\/li>\r\n<\/ul>\r\n<p class=\"p5\"><i>Exemple:<\/i> Un assistent digital que llegeix documents, extrau informaci\u00f3 i escriu un resum adaptat al teu estil i objectius.<\/p>\r\n\r\n<h4 class=\"p6\"><b>Difer\u00e8ncies clau<\/b><\/h4>\r\n<table>\r\n<thead>\r\n<tr>\r\n<th>\r\n<p class=\"p1\"><b>Aspecte<\/b><\/p>\r\n<\/th>\r\n<th>\r\n<p class=\"p1\"><b>Agent cl\u00e0ssic<\/b><\/p>\r\n<\/th>\r\n<th>\r\n<p class=\"p1\"><b>Agent LLM<\/b><\/p>\r\n<\/th>\r\n<\/tr>\r\n<\/thead>\r\n<tbody>\r\n<tr>\r\n<td>\r\n<p class=\"p1\"><b>Entrades<\/b><\/p>\r\n<\/td>\r\n<td>\r\n<p class=\"p1\">Dades estructurades<\/p>\r\n<\/td>\r\n<td>\r\n<p class=\"p1\">Llenguatge natural, textos, APIs<\/p>\r\n<\/td>\r\n<\/tr>\r\n<tr>\r\n<td>\r\n<p class=\"p1\"><b>Decisi\u00f3<\/b><\/p>\r\n<\/td>\r\n<td>\r\n<p class=\"p1\">Regles i l\u00f2gica programada<\/p>\r\n<\/td>\r\n<td>\r\n<p class=\"p1\">Raonament estad\u00edstic i context<\/p>\r\n<\/td>\r\n<\/tr>\r\n<tr>\r\n<td>\r\n<p class=\"p1\"><b>Adaptabilitat<\/b><\/p>\r\n<\/td>\r\n<td>\r\n<p class=\"p1\">Baixa<\/p>\r\n<\/td>\r\n<td>\r\n<p class=\"p1\">Molt alta<\/p>\r\n<\/td>\r\n<\/tr>\r\n<tr>\r\n<td>\r\n<p class=\"p1\"><b>Coneixement<\/b><\/p>\r\n<\/td>\r\n<td>\r\n<p class=\"p1\">Tancat i expl\u00edcit<\/p>\r\n<\/td>\r\n<td>\r\n<p class=\"p1\">Ampl\u00edssim i impl\u00edcit<\/p>\r\n<\/td>\r\n<\/tr>\r\n<tr>\r\n<td>\r\n<p class=\"p1\"><b>Context<\/b><\/p>\r\n<\/td>\r\n<td>\r\n<p class=\"p1\">Limitat<\/p>\r\n<\/td>\r\n<td>\r\n<p class=\"p1\">Pot gestionar contextos llargs i canviants<\/p>\r\n<\/td>\r\n<\/tr>\r\n<tr>\r\n<td>\r\n<p class=\"p1\"><b>Creativitat<\/b><\/p>\r\n<\/td>\r\n<td>\r\n<p class=\"p1\">Inexistent<\/p>\r\n<\/td>\r\n<td>\r\n<p class=\"p1\">Pot improvisar, sintetitzar i adaptar<\/p>\r\n<\/td>\r\n<\/tr>\r\n<\/tbody>\r\n<\/table>\r\n\r\n<hr \/>\r\n\r\n<h4 class=\"p3\"><b>Quin \u00e9s millor?<\/b><b><\/b><\/h4>\r\n<p class=\"p4\">Dep\u00e8n de l\u2019\u00fas.<\/p>\r\n<p class=\"p4\"><span class=\"s2\"><b>Agents cl\u00e0ssics<\/b><\/span> s\u00f3n ideals per a entorns <span class=\"s2\"><b>molt definits i predictibles<\/b><\/span>, on la fiabilitat \u00e9s clau (per exemple, un sistema de control industrial).<\/p>\r\n<p class=\"p4\"><span class=\"s2\"><b>Agents LLM<\/b><\/span> brillen en <span class=\"s2\"><b>entorns oberts, canviants i amb dades no estructurades<\/b><\/span>, com atenci\u00f3 al client, assistents personals, automatitzaci\u00f3 intel\u00b7ligent o consultoria d\u2019informaci\u00f3.<\/p>\r\n<p class=\"p4\">En molts casos, la millor opci\u00f3 \u00e9s <span class=\"s2\"><b>combinar-los<\/b><\/span>: fer servir agents LLM per la part flexible i natural, i agents cl\u00e0ssics per a l\u2019execuci\u00f3 rigorosa i controlada.<\/p>\r\n\r\n<h4 class=\"p3\"><b>Cap on va el futur?<\/b><b><\/b><\/h4>\r\n<p class=\"p4\">La tend\u00e8ncia clara \u00e9s cap a agents <span class=\"s2\"><b>m\u00e9s h\u00edbrids i capa\u00e7os<\/b><\/span>:<\/p>\r\n\r\n<ul>\r\n \t<li class=\"p5\">LLMs que aprenen a controlar sistemes amb regles.<\/li>\r\n \t<li class=\"p5\">Agents cl\u00e0ssics que incorporen m\u00f2duls de llenguatge.<\/li>\r\n \t<li class=\"p5\">Sistemes multi-agent on <a href=\"https:\/\/elink.cat\/blog\/quan-els-agents-collaboren-el-poder-dels-sistemes-multi-agent\/\"><span class=\"s2\"><b>cada tipus d\u2019agent t\u00e9 un rol espec\u00edfic<\/b><\/span> <\/a>(ja en vaig parlar en l&#8217;article anterior!).<\/li>\r\n<\/ul>\r\n<p class=\"p6\"><span class=\"s3\">El que abans era nom\u00e9s l\u00f2gica, ara \u00e9s <\/span><b>una conversa constant entre raonament i acci\u00f3<\/b><span class=\"s3\">.<\/span><\/p>\r\n<p class=\"p4\">Els <span class=\"s2\"><b>agents cl\u00e0ssics<\/b><\/span> ens han portat molt lluny en entorns estructurats. Per\u00f2 els <span class=\"s2\"><b>agents basats en LLM<\/b><\/span> han obert una nova era d\u2019intel\u00b7lig\u00e8ncia adaptativa, capa\u00e7 de gestionar informaci\u00f3 complexa, parlar amb humans i actuar amb criteri.<\/p>\r\n<p class=\"p4\">El futur de la IA no \u00e9s un o l\u2019altre, sin\u00f3 una <span class=\"s2\"><b>col\u00b7laboraci\u00f3 intel\u00b7ligent<\/b><\/span> entre l\u00f2gica i llenguatge, entre regles i raonament.<\/p>\n\n\n<div class=\"wp-block-group elink-related\"><div class=\"wp-block-group__inner-container is-layout-flow wp-block-group-is-layout-flow\">\n<h3 class=\"wp-block-heading\">Per seguir llegint<\/h3>\n\n\n\n<ul class=\"wp-block-list\"><li><a href=\"https:\/\/elink.cat\/blog\/que-es-un-agent-dia-i-com-pensen\/\">Qu\u00e8 \u00e9s un agent d&#8217;IA? I com \u00abpensen\u00bb<\/a><\/li><li><a href=\"https:\/\/elink.cat\/blog\/com-funciona-un-agent-dia-arquitectura-i-components-clau\/\">Com funciona un agent d\u2019IA? Arquitectura i components clau<\/a><\/li><\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">Si vols que el teu equip ho domini, mira la nostra <a href=\"https:\/\/elink.cat\/serveis\/formacio\">Formaci\u00f3 en IA per a equips<\/a>.<\/p>\n<\/div><\/div>\n","protected":false},"excerpt":{"rendered":"<p><span class=\"span-reading-time rt-reading-time\" style=\"display: block;\"><span class=\"rt-label rt-prefix\">Temps de lectura: <\/span> <span class=\"rt-time\"> 2<\/span> <span class=\"rt-label rt-postfix\">minuts<\/span><\/span>Com tots ja sabeu, el m\u00f3n dels agents d\u2019intel\u00b7lig\u00e8ncia artificial ha viscut una revoluci\u00f3 amb l\u2019aparici\u00f3 dels LLMs (Large Language Models). Per\u00f2 abans que aquests models entressin en escena, ja<\/p>\n","protected":false},"author":1,"featured_media":2080,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"breadcrumbs_single_post":"","page_title_panel":"","breadcrumbs_single_page":"","single_page_alignment":"","single_page_margin":"","page_structure_type":"","content_style_source":"","content_style":"","blog_post_streched_ed":"","blog_page_streched_ed":"","has_transparent_header":"","disable_transparent_header":"","vertical_spacing_source":"","content_area_spacing":"","single_post_content_background":"","single_page_content_background":"","single_post_boxed_content_spacing":"","single_page_boxed_content_spacing":"","single_post_content_boxed_radius":"","single_page_content_boxed_radius":"","disable_featured_image":"","disable_post_tags":"","disable_author_box":"","disable_posts_navigation":"","disable_comments":"","disable_related_posts":"","disable_header":"","disable_footer":"","footnotes":""},"categories":[14],"tags":[],"class_list":["post-2040","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-tendencies","rishi-post"],"_links":{"self":[{"href":"https:\/\/elink.cat\/blog\/wp-json\/wp\/v2\/posts\/2040","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/elink.cat\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/elink.cat\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/elink.cat\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/elink.cat\/blog\/wp-json\/wp\/v2\/comments?post=2040"}],"version-history":[{"count":3,"href":"https:\/\/elink.cat\/blog\/wp-json\/wp\/v2\/posts\/2040\/revisions"}],"predecessor-version":[{"id":2929,"href":"https:\/\/elink.cat\/blog\/wp-json\/wp\/v2\/posts\/2040\/revisions\/2929"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/elink.cat\/blog\/wp-json\/wp\/v2\/media\/2080"}],"wp:attachment":[{"href":"https:\/\/elink.cat\/blog\/wp-json\/wp\/v2\/media?parent=2040"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/elink.cat\/blog\/wp-json\/wp\/v2\/categories?post=2040"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/elink.cat\/blog\/wp-json\/wp\/v2\/tags?post=2040"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}