AI Agentic Workflows – Class 3 (DO Framework)
1. Stochasticity Problem
AI har step par 90% accurate ho sakta hai.
Lekin multiple steps mein accuracy girti hai.
Example:
90% × 90% × 90% × 90% × 90%
≈ 59% overall success rate
Isi problem ko solve karne ke liye DO Framework use hota hai.
2. DO Framework (3-Layer Architecture)
A. Directive
Natural language document hota hai.
Isme define hota hai:
Purpose
Inputs
Process
Expected Output
B. Orchestration
AI ka brain hota hai.
Directive ko read karta hai.
Different execution scripts ko connect aur manage karta hai.
C. Execution
Deterministic code/scripts hote hain.
Repeated tasks ko 100% reliable banate hain.
3. System Prompt (agents.md)
Agent ko initial instructions deta hai.
Ship navigation ki tarah direction set karta hai.
Agent ko batata hai kya karna hai aur kaise karna hai.
4. Self-Annealing
Agent apni mistakes se improve karta hai:
Errors read karta hai.
Code rewrite karta hai.
Tools ko better banata hai.
Performance continuously improve hoti hai.
5. Minecraft Analogy
Bare hands se kaam slow hota hai.
Better tools (Diamond tools) se speed aur efficiency badhti hai.
Execution scripts bhi AI ke liye advanced tools ki tarah kaam karte hain.
6. Caveman Analogy
Spear ke bina survival ≈ 20%
Spear ke saath survival ≈ 99%
Isi tarah execution scripts AI ki reliability ko bahut improve karte hain.
7. Final Outcome
DO Framework ki wajah se:
Reliability ≈ 99%
AI zyada accurate hota hai.
Multi-step workflows stable hote hain.
Complex tasks efficiently complete hote hain.
Exam/Interview Points
Stochasticity Problem: Multi-step accuracy drop.
DO = Directive + Orchestration + Execution
Directive: Instructions.
Orchestration: AI Brain.
Execution: Reliable Code.
System Prompt: Initial guidance.
Self-Annealing: Self-improvement.
Goal: 99% reliable AI agents.