Lesson · Hinglish · ⏱ 8 min padhne mein
🤖 AI agents: plan, act, observe, repeat
EnglishHinglish
🏠 Asli zindagi ki misaal
Goa trip plan karte waqt aap flight prices dekhte ho, phir beach ke paas hotels, phir notice karte ho ki hotel airport se door hai, to doosre airport ki flights dobara check karte ho, aur phir book karte ho. Aap plan karte ho, act karte ho, result dekhte ho aur adjust karte ho. AI agent bhi isi loop mein kaam karta hai, har step pe tools use karke, jab tak task poora na ho.
💡 Aasaan shabdon mein
Agent loop mein chalta hua LLM hai: woh agla action decide karta hai, tool call karta hai, result padhta hai, aur tab tak repeat karta hai jab tak final jawaab na de sake ya limit na aa jaaye. Agents ko memory (ab tak kya hua), saaf tools, aur rukne ke rules chahiye, jaise maximum steps aur budget. LangGraph ya agent SDKs jaise frameworks loop aur state sambhaalte hain. Agents naye tareekon se fail hote hain: hamesha loop karte rehte hain, galat tool chun lete hain, ya kharab data pe action le lete hain. Fixed workflow se shuru karo, autonomy sirf wahan jodo jahan saaf faayda ho, aur high-impact steps pe insaan ki approval rakho.
🎬 Dekho kaise kaam karta hai
Agent loop mein chalta hai: agla step socho, tool se act karo, result dekho, aur repeat karo, jab tak goal poora na ho ya step limit na rok de.
🪜 Step by step
- Agent ko ek saaf goal aur kuch tools do.
- Loop: model agla step plan karta hai, tool call karta hai, result padhta hai, aur phir decide karta hai.
- Limits lagao: maximum steps, maximum cost aur timeout, taaki hamesha loop na chale.
- Payments, customers ko emails ya data delete jaise risky actions se pehle human approval zaroori karo.
- Har step log karo (trace) taaki dekh sako usne jo kiya kyun kiya, aur realistic tasks pe evaluate karo.
🔑 Zaroori shabd
- Agent loop
- Socho → act (tool) → dekho → repeat.
- State / memory
- Task mein agent ne ab tak kya kiya aur kya seekha.
- Guardrails
- Steps, cost aur actions pe limits jo agent ko safe rakhti hain.
👀 Example dekho
Goal: "Find 3 well-rated cafés near Koramangala open after 10 pm"
Step 1 think → search_places("cafés Koramangala")
Step 2 observe → 12 results
Step 3 think → get_hours() for each, filter rating ≥ 4.2
Step 4 observe → 3 match
Step 5 → final answer with names, ratings, closing times ✅ (max_steps = 8)Har step pichhle result ko use karta hai. Step limit endless loops ko rokti hai.
⚠️ Aam galtiyaan jinse bachna hai
- Jahan fixed workflow kaam kare wahan agent lagana. Fixed steps saste, tez aur test karne mein aasaan hain.
- Rukne ke rules na hona, jisse loops bhaagte rehte hain aur bill bada hota hai.
- Web pages ya documents ko agent ko instructions dene dena (prompt injection). Unhe data maano.
✅ Jaldi se check karo
Agent same search tool 50 baar call karta ja raha hai. Isse kya rokna chahiye tha?
- Zyaada temperature
- Bada model
- Max-steps limit aur loop detection jaise guardrails
Jawaab dekho
C. Max-steps limit aur loop detection jaise guardrails: Agents ko saaf rukne ke rules chahiye. Sirf model ke bharose mat raho.
🙋 Seekhne wale yeh poochhte hain
Agent aur chatbot mein kya fark hai?
Chatbot zyaadatar jawaab deta hai. Agent kai steps mein tools se actions leta hai taaki kaam poora ho.
Kya LangGraph jaisa framework chahiye?
Shuru mein nahi. Apne code mein simple loop sabse zyaada sikhata hai. Frameworks baad mein state, retries aur multi step flows mein madad karte hain.
📌 Yaad rakho
Agent ka formula: LLM + tools + loop + memory + guardrails. Inmein se ek bhi kam, to agent bharosemand nahi.
📖 Is lesson ke shabd (AI dictionary)
Poora course, aapke roadmap ke hisaab se
Progress tracking, quick checks aur step-by-step order ke saath. Yeh lesson in courses ka hissa hai:
Course mein seekho