{"id":11340,"date":"2026-10-04T00:25:12","date_gmt":"2026-10-04T00:25:12","guid":{"rendered":"https:\/\/bitep.net\/blog\/2026\/10\/04\/infoworldcom\/"},"modified":"2026-10-04T00:25:12","modified_gmt":"2026-10-04T00:25:12","slug":"infoworldcom","status":"publish","type":"post","link":"https:\/\/bitep.net\/blog\/2026\/10\/04\/infoworldcom\/","title":{"rendered":"infoworld.com"},"content":{"rendered":"<h2>S\u00fcni \u0130ntellektd\u0259 Yeni Era: Sad\u0259 \u00c7atbotlardan &#8220;Agentik AI&#8221; v\u0259 Avtonom Sisteml\u0259r\u0259 Ke\u00e7id<\/h2>\n<p>S\u00fcni intellekt (S\u0130) s\u0259nayesi s\u00fcr\u0259tli v\u0259 k\u00f6kl\u00fc bir transformasiya m\u0259rh\u0259l\u0259sin\u0259 daxil olub. \u018fg\u0259r son iki il \u0259rzind\u0259 \u0259sas diqq\u0259t b\u00f6y\u00fck dil modell\u0259rinin (LLM) t\u0259k bir sor\u011fuya (prompt) cavab olaraq m\u0259tn, \u015f\u0259kil v\u0259 ya kod generasiya etm\u0259sin\u0259 y\u00f6n\u0259lmi\u015fdis\u0259, art\u0131q bu yana\u015fma \u00f6z yerini <strong>&#8220;Agentic AI&#8221; (Agentik S\u00fcni \u0130ntellekt)<\/strong> v\u0259 davaml\u0131 i\u015f ax\u0131nlar\u0131na (persistent workflows) verir. \u0130stifad\u0259\u00e7il\u0259rin yaln\u0131z sual verib cavab ald\u0131\u011f\u0131 passiv \u00e7atbotlar arxa plana ke\u00e7ir; onlar\u0131n yerini m\u00fcst\u0259qil q\u0259rarlar q\u0259bul ed\u0259 bil\u0259n, al\u0259tl\u0259rd\u0259n istifad\u0259 ed\u0259n v\u0259 m\u00fcr\u0259kk\u0259b tap\u015f\u0131r\u0131qlar\u0131 add\u0131m-add\u0131m icra ed\u0259n avtonom agentl\u0259r tutur.<\/p>\n<p>Bu texnoloji s\u0131\u00e7ray\u0131\u015f\u0131n arxas\u0131nda modell\u0259rin t\u0259kc\u0259 m\u0259lumat\u0131 xat\u0131rlamas\u0131 deyil, h\u0259m d\u0259 \u00e7oxm\u0259rh\u0259l\u0259li m\u0259ntiqi m\u00fchakim\u0259 (advanced reasoning) y\u00fcr\u00fctm\u0259k v\u0259 \u00f6z s\u0259hvl\u0259rini d\u00fcz\u0259ltm\u0259k bacar\u0131\u011f\u0131 dayan\u0131r.<\/p>\n<h2>Agentik \u0130\u015f Ax\u0131nlar\u0131n\u0131n (Agentic Workflows) \u018fsas Mexanizmi<\/h2>\n<p>Klassik \u00e7atbotlar x\u0259tti v\u0259 bird\u0259f\u0259lik sor\u011fu-cavab arxitekturas\u0131 \u00fcz\u0259rind\u0259 i\u015fl\u0259yir (zero-shot v\u0259 ya few-shot prompting). Model daxil edil\u0259n kontekst \u0259sas\u0131nda \u0259n uy\u011fun ehtimal olunan tokenl\u0259ri generasiya edir v\u0259 proses orada dayan\u0131r. Lakin agentik i\u015f ax\u0131nlar\u0131nda proses tsiklik v\u0259 adaptivdir. Sistem qar\u015f\u0131ya qoyulan \u00fcmumi m\u0259qs\u0259di alt tap\u015f\u0131r\u0131qlara b\u00f6l\u00fcr, h\u0259r bir add\u0131m\u0131n n\u0259tic\u0259sini qiym\u0259tl\u0259ndirir v\u0259 laz\u0131m g\u0259ldikd\u0259 xarici m\u00fchitl\u0259 \u0259laq\u0259 qurur.<\/p>\n<p>Agentik sistemin \u0259sas komponentl\u0259ri a\u015fa\u011f\u0131dak\u0131lard\u0131r:<\/p>\n<ul>\n<li><strong>Planla\u015fd\u0131rma (Planning):<\/strong> Model tap\u015f\u0131r\u0131\u011f\u0131 h\u0259ll etm\u0259k \u00fc\u00e7\u00fcn ard\u0131c\u0131l add\u0131mlar strategiyas\u0131 formala\u015fd\u0131r\u0131r (m\u0259s\u0259l\u0259n, Chain-of-Thought v\u0259 ya Tree-of-Thoughts yana\u015fmalar\u0131).<\/li>\n<li><strong>Yadda\u015f (Memory):<\/strong> Q\u0131sam\u00fcdd\u0259tli (i\u015fl\u0259k kontekst) v\u0259 uzunm\u00fcdd\u0259tli (vektor bazalar\u0131 &#8211; RAG) yadda\u015f vasit\u0259sil\u0259 agent \u0259vv\u0259lki add\u0131mlarda \u0259ld\u0259 etdiyi n\u0259tic\u0259l\u0259ri xat\u0131rlay\u0131r.<\/li>\n<li><strong>Al\u0259tl\u0259rd\u0259n \u0130stifad\u0259 (Tool Use \/ Function Calling):<\/strong> Agentl\u0259r kalkulyatorlardan, API-l\u0259rd\u0259n, veril\u0259nl\u0259r bazalar\u0131ndan v\u0259 ya brauzerl\u0259rd\u0259n istifad\u0259 ed\u0259r\u0259k real vaxtda m\u0259lumat toplay\u0131r v\u0259 \u0259m\u0259liyyat icra edir.<\/li>\n<li><strong>\u018fks-\u0259laq\u0259 v\u0259 \u0130cra (Action &#038; Reflection):<\/strong> \u018fld\u0259 edil\u0259n aral\u0131q n\u0259tic\u0259 t\u0259hlil edilir v\u0259 strategiyaya d\u00fcz\u0259li\u015fl\u0259r edilir.<\/li>\n<\/ul>\n<h2>\u00d6z\u00fcn\u00fc Do\u011frulama (Self-Verification) D\u00f6vr\u0259l\u0259ri: Hall\u00fcsinasiyalara Son<\/h2>\n<p>B\u00f6y\u00fck dil modell\u0259rinin \u0259n z\u0259if n\u00f6qt\u0259si h\u0259mi\u015f\u0259 &#8220;hall\u00fcsinasiya&#8221; \u2014 y\u0259ni modelin faktiki olaraq yanl\u0131\u015f, lakin inand\u0131r\u0131c\u0131 g\u00f6r\u00fcn\u0259n m\u0259lumatlar uydurmas\u0131 olub. Agentik arxitekturan\u0131n \u0259n b\u00f6y\u00fck texnoloji nailiyy\u0259tl\u0259rind\u0259n biri <strong>daxili \u00f6z\u00fcn\u00fc do\u011frulama (self-verification)<\/strong> d\u00f6vr\u0259l\u0259rinin t\u0259tbiqidir.<\/p>\n<p>Bu mexanizmd\u0259 model cavab\u0131 istifad\u0259\u00e7iy\u0259 t\u0259qdim etm\u0259zd\u0259n \u0259vv\u0259l daxili audit apar\u0131r:<\/p>\n<ul>\n<li><strong>Riyazi D\u0259qiqlik:<\/strong> \u018fg\u0259r cavab riyazi hesablamalara \u0259saslan\u0131rsa, model hesablaman\u0131 Python m\u00fchitind\u0259 simulyasiya ed\u0259r\u0259k v\u0259 ya formal m\u0259ntiq yoxlay\u0131c\u0131lar\u0131 il\u0259 t\u0259krar yoxlay\u0131r.<\/li>\n<li><strong>Kodun Yoxlanmas\u0131:<\/strong> Generasiya edil\u0259n proqram kodu sandbox (t\u0259crid olunmu\u015f virtual m\u00fchit) daxilind\u0259 icra edilir, sintaksis v\u0259 icra x\u0259talar\u0131 (runtime errors) a\u015fkarland\u0131qda model kodu avtonom \u015f\u0259kild\u0259 yenid\u0259n yaz\u0131r.<\/li>\n<li><strong>M\u0259ntiqi Tutarl\u0131l\u0131q:<\/strong> Model ilkin \u015f\u0259rtl\u0259rl\u0259 son n\u0259tic\u0259 aras\u0131ndak\u0131 m\u0259ntiqi ziddiyy\u0259tl\u0259ri axtar\u0131r v\u0259 s\u00fcbut z\u0259ncirini t\u0259hlil edir.<\/li>\n<\/ul>\n<p>Bu yana\u015fma insan m\u00fcdaxil\u0259sin\u0259 (human-in-the-loop) ehtiyac\u0131 k\u0259skin d\u0259r\u0259c\u0259d\u0259 azald\u0131r v\u0259 s\u00fcni intellektin kritik sektorlarda (maliyy\u0259, tibbi diaqnostika, proqram t\u0259minat\u0131 m\u00fch\u0259ndisliyi) etibarl\u0131l\u0131\u011f\u0131n\u0131 art\u0131r\u0131r.<\/p>\n<h2>M\u00fcqayis\u0259: \u018fn\u0259n\u0259vi LLM \u00c7atbotu v\u0259 Agentik AI<\/h2>\n<table border=\"1\" cellpadding=\"8\" cellspacing=\"0\" >\n<thead>\n<tr >\n<th>X\u00fcsusiyy\u0259t<\/th>\n<th>\u018fn\u0259n\u0259vi Generativ \u00c7atbot<\/th>\n<th>Agentik AI (\u0130\u015f Ax\u0131nlar\u0131)<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><strong>\u0130\u015fl\u0259m\u0259 \u015e\u0259kli<\/strong><\/td>\n<td>T\u0259k sor\u011fu &#8211; t\u0259k cavab (Single-turn)<\/td>\n<td>Davaml\u0131, \u00e7oxm\u0259rh\u0259l\u0259li d\u00f6vr\u0259 (Iterative loops)<\/td>\n<\/tr>\n<tr>\n<td><strong>X\u0259ta \u0130dar\u0259etm\u0259si<\/strong><\/td>\n<td>\u0130stifad\u0259\u00e7i s\u0259hvi g\u00f6r\u00fcb yeni prompt yazmal\u0131d\u0131r<\/td>\n<td>Daxili self-verification il\u0259 \u00f6z s\u0259hvini a\u015fkarlay\u0131r v\u0259 d\u00fcz\u0259ldir<\/td>\n<\/tr>\n<tr>\n<td><strong>Xarici D\u00fcnya \u018flaq\u0259si<\/strong><\/td>\n<td>M\u0259hdud v\u0259 ya passiv (yaln\u0131z oxuma)<\/td>\n<td>Aktiv al\u0259t istifad\u0259si (API \u00e7a\u011f\u0131rma, kod icras\u0131, fayl oxuma\/yazma)<\/td>\n<\/tr>\n<tr>\n<td><strong>Hall\u00fcsinasiya Ehtimal\u0131<\/strong><\/td>\n<td>Y\u00fcks\u0259k (statik ehtimal paylanmas\u0131)<\/td>\n<td>\u00c7ox a\u015fa\u011f\u0131 (fakt v\u0259 icra yoxlamalar\u0131 say\u0259sind\u0259)<\/td>\n<\/tr>\n<tr>\n<td><strong>Avtonomiya S\u0259viyy\u0259si<\/strong><\/td>\n<td>A\u015fa\u011f\u0131 (\u0130stifad\u0259\u00e7i t\u0259r\u0259find\u0259n idar\u0259 olunur)<\/td>\n<td>Y\u00fcks\u0259k (H\u0259d\u0259f verilir, icra agent\u0259 h\u0259val\u0259 edilir)<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<h2>Praktiki B\u0259l\u0259d\u00e7i: Python v\u0259 LangChain il\u0259 Sad\u0259 Yoxlama Agentinin Qurulmas\u0131<\/h2>\n<p>Agentik i\u015f ax\u0131nlar\u0131n\u0131 v\u0259 \u00f6z\u00fcn\u00fc yoxlama d\u00f6vr\u0259sini anlamaq \u00fc\u00e7\u00fcn Python m\u00fchitind\u0259 sad\u0259 bir arxitektura quraq. Bu n\u00fcmun\u0259d\u0259 agent veril\u0259n problemi h\u0259ll etm\u0259k \u00fc\u00e7\u00fcn kod yazacaq, onu daxili m\u00fchitd\u0259 yoxlayacaq v\u0259 x\u0259ta ba\u015f ver\u0259rs\u0259 \u00f6z\u00fcn\u0259 d\u00fcz\u0259li\u015f ed\u0259c\u0259k.<\/p>\n<h3>1. T\u0259l\u0259b Olunan Kitabxanalar\u0131n Qura\u015fd\u0131r\u0131lmas\u0131<\/h3>\n<pre><code class=\"language-bash\">pip install langchain langchain-openai langchain-experimental<\/code><\/pre>\n<h3>2. Agentik Kod Yoxlama Skripti<\/h3>\n<p>A\u015fa\u011f\u0131dak\u0131 Python n\u00fcmun\u0259si agentin x\u0259tan\u0131 a\u015fkar edib \u00f6z\u00fcn\u00fc d\u00fcz\u0259ltm\u0259 prinsipini n\u00fcmayi\u015f etdirir:<\/p>\n<pre><code class=\"language-bash\">cat &lt;&lt; 'EOF' &gt; agent_loop.py\nimport sys\nfrom langchain_openai import ChatOpenAI\nfrom langchain.schema import SystemMessage, HumanMessage\n\nllm = ChatOpenAI(model=\"gpt-4o\", temperature=0)\n\ndef execute_code(code_string):\n    \"\"\"Kodu t\u0259hl\u00fck\u0259siz \u015f\u0259kild\u0259 icra edir v\u0259 \u00e7\u0131x\u0131\u015f\u0131 v\u0259 ya x\u0259tan\u0131 qaytar\u0131r.\"\"\"\n    try:\n        local_scope = {}\n        exec(code_string, {}, local_scope)\n        return True, local_scope.get(\"result\", \"Kod u\u011furla icra olundu.\")\n    except Exception as e:\n        return False, str(e)\n\ntask = \"1-d\u0259n 100-\u0259 q\u0259d\u0259r olan sad\u0259 \u0259d\u0259dl\u0259rin c\u0259mini tapan Python funksiyas\u0131 yaz v\u0259 n\u0259tic\u0259ni 'result' d\u0259yi\u015f\u0259nin\u0259 m\u0259nims\u0259t.\"\nsystem_prompt = \"Yaln\u0131z icra oluna bil\u0259n Python kodu yaz. Kod bloklar\u0131n\u0131 python format\u0131nda t\u0259qdim etm\u0259, birba\u015fa xam kod g\u00f6nd\u0259r.\"\n\n# \u0130lkin c\u0259hd\nresponse = llm.invoke([SystemMessage(content=system_prompt), HumanMessage(content=task)])\ncode = response.content\n\n# Self-Verification Loop (\u00d6z\u00fcn\u00fc Do\u011frulama D\u00f6vr\u0259si)\nmax_retries = 3\nfor attempt in range(max_retries):\n    success, output = execute_code(code)\n    if success:\n        print(f\"[U\u011eURLU] N\u0259tic\u0259 tap\u0131ld\u0131: {output}\")\n        break\n    else:\n        print(f\"[X\u018fTA {attempt+1}]: {output}. D\u00fcz\u0259li\u015f edilir...\")\n        correction_prompt = f\"Yazd\u0131\u011f\u0131n kod bu x\u0259tan\u0131 verdi: {output}. Z\u0259hm\u0259t olmasa kodu d\u00fcz\u0259lt:\\n{code}\"\n        response = llm.invoke([SystemMessage(content=system_prompt), HumanMessage(content=correction_prompt)])\n        code = response.content\nEOF\npython3 agent_loop.py<\/code><\/pre>\n<h2>S\u0259nayey\u0259 v\u0259 T\u0259rtibat\u00e7\u0131lara T\u0259siri<\/h2>\n<p>Agentik s\u00fcni intellekt\u0259 ke\u00e7id \u0130T v\u0259 biznes prosesl\u0259rind\u0259 paradiqman\u0131 tamamil\u0259 d\u0259yi\u015fir:<\/p>\n<ul>\n<li><strong>Proqram T\u0259minat\u0131n\u0131n \u0130nki\u015faf\u0131:<\/strong> T\u0259rtibat\u00e7\u0131lar art\u0131q sad\u0259c\u0259 kod yazan assistentl\u0259rl\u0259 deyil, repozitoriyan\u0131 klonlay\u0131b, arxitekturan\u0131 analiz ed\u0259n, testl\u0259ri yazan v\u0259 CI\/CD konveyerl\u0259rini m\u00fcst\u0259qil sazlayan avtonom m\u00fch\u0259ndisl\u0259rl\u0259 (m\u0259s\u0259l\u0259n, Devin tipli sisteml\u0259r) i\u015fl\u0259yirl\u0259r.<\/li>\n<li><strong>Gecikm\u0259 (Latency) v\u0259 D\u0259qiqlik Balans\u0131:<\/strong> Agentik i\u015f ax\u0131nlar\u0131 bird\u0259f\u0259lik sor\u011fulardan daha \u00e7ox vaxt apar\u0131r (b\u0259z\u0259n d\u0259qiq\u0259l\u0259rl\u0259 davam ed\u0259n hesablama d\u00f6vr\u0259l\u0259ri), lakin \u0259ld\u0259 olunan n\u0259tic\u0259nin d\u0259qiqliyi v\u0259 d\u0259y\u0259ri t\u0259k cavabl\u0131 sisteml\u0259rd\u0259n qat-qat y\u00fcks\u0259kdir.<\/li>\n<li><strong>Sistem T\u0259hl\u00fck\u0259sizliyi v\u0259 N\u0259zar\u0259t:<\/strong> Agentl\u0259r\u0259 al\u0259tl\u0259rd\u0259n istifad\u0259 v\u0259 sistem \u0259mrl\u0259rini icra etm\u0259k s\u0259lahiyy\u0259ti verildiyi \u00fc\u00e7\u00fcn &#8220;Prompt Injection&#8221; v\u0259 arzuolunmaz \u0259m\u0259liyyatlar\u0131n qar\u015f\u0131s\u0131n\u0131 almaq \u00fc\u00e7\u00fcn t\u0259hl\u00fck\u0259sizlik baryerl\u0259ri (Guardrails) \u0259sas prioritet\u0259 \u00e7evrilir.<\/li>\n<\/ul>\n<h2>BITEP N\u0259tic\u0259si<\/h2>\n<p>Sad\u0259 dil modell\u0259ri art\u0131q texnoloji bazad\u0131r; innovasiyan\u0131n n\u00f6vb\u0259ti dal\u011fas\u0131 bu modell\u0259rin \u00fcz\u0259rind\u0259 qurulan agentik arxitekturalard\u0131r. \u00d6z\u00fcn\u00fc yoxlama (self-verification), al\u0259t inteqrasiyas\u0131 v\u0259 davaml\u0131 i\u015f ax\u0131nlar\u0131 s\u00fcni intellekti passiv m\u0259tn generatorundan \u00e7\u0131xar\u0131b, real probleml\u0259ri m\u00fcst\u0259qil \u015f\u0259kild\u0259 h\u0259ll ed\u0259n r\u0259q\u0259msal i\u015f\u00e7i q\u00fcvv\u0259sin\u0259 \u00e7evirir. T\u0259rtibat\u00e7\u0131lar v\u0259 m\u00fch\u0259ndisl\u0259r \u00fc\u00e7\u00fcn bu d\u0259yi\u015fikliyi anlamaq v\u0259 t\u0259tbiql\u0259rini agentik yana\u015fma il\u0259 yenid\u0259n dizayn etm\u0259k art\u0131q se\u00e7im deyil, z\u0259rur\u0259tdir.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>S\u00fcni \u0130ntellektd\u0259 Yeni Era: Sad\u0259 \u00c7atbotlardan &#8220;Agentik AI&#8221; v\u0259 Avtonom Sisteml\u0259r\u0259 Ke\u00e7id S\u00fcni intellekt (S\u0130) s\u0259nayesi s\u00fcr\u0259tli v\u0259 k\u00f6kl\u00fc bir transformasiya m\u0259rh\u0259l\u0259sin\u0259 daxil olub. \u018fg\u0259r son iki il \u0259rzind\u0259 \u0259sas diqq\u0259t b\u00f6y\u00fck dil modell\u0259rinin (LLM) t\u0259k bir sor\u011fuya (prompt) cavab olaraq m\u0259tn, \u015f\u0259kil v\u0259 ya kod generasiya etm\u0259sin\u0259 y\u00f6n\u0259lmi\u015fdis\u0259, art\u0131q bu yana\u015fma \u00f6z yerini &#8220;Agentic [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":11339,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"om_disable_all_campaigns":false,"_monsterinsights_skip_tracking":false,"_jetpack_memberships_contains_paid_content":false,"footnotes":""},"categories":[1],"tags":[],"class_list":["post-11340","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-uncategorized"],"aioseo_notices":[],"jetpack_featured_media_url":"https:\/\/bitep.net\/blog\/wp-content\/uploads\/2026\/10\/bitep-ai-1791073507318-7.png","jetpack_sharing_enabled":true,"_links":{"self":[{"href":"https:\/\/bitep.net\/blog\/wp-json\/wp\/v2\/posts\/11340","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/bitep.net\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/bitep.net\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/bitep.net\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/bitep.net\/blog\/wp-json\/wp\/v2\/comments?post=11340"}],"version-history":[{"count":0,"href":"https:\/\/bitep.net\/blog\/wp-json\/wp\/v2\/posts\/11340\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/bitep.net\/blog\/wp-json\/wp\/v2\/media\/11339"}],"wp:attachment":[{"href":"https:\/\/bitep.net\/blog\/wp-json\/wp\/v2\/media?parent=11340"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/bitep.net\/blog\/wp-json\/wp\/v2\/categories?post=11340"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/bitep.net\/blog\/wp-json\/wp\/v2\/tags?post=11340"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}