This article presents the AgentFlow frameworka new paradigm of no-code programming built on natural-language structural pseudocode, executed through Large Language Models (LLMs). The framework utilizes logical primitives-Condition, Loop, Function, Label, Gotoas fundamental building blocks for constructing complex agent behaviors. An agent model is introduced, where each agent has: a role, state, logic, and communication capability. The concept of a swarm of agents is proposed, enabling LLMs to simulate parallel task execution and process complex queries without writing code. The article also describes a memory system for self-learning, provides examples of data analysis, and outlines formal rules for agent-to-agent interaction. Keywords: No-Code Programming, Large Language Models (LLM), Agent Model, Swarm of Agents, Logical Primitives, Parallel Execution Via Prompts, Agent Self-Learning