Agentic Framework

An agentic framework is a software library or platform that provides the building blocks for creating AI agents — autonomous systems that can plan their own steps, use tools, and complete complex tasks without a human guiding every action. Instead of writing all the orchestration logic from scratch, developers use a framework to handle the common patterns.

What frameworks actually do

Building an AI agent from scratch requires solving several hard problems: how does the agent decide what to do next? How does it call external tools? How does it handle errors? How does it remember what it has already done? How do multiple agents coordinate?

An agentic framework provides pre-built solutions for these problems:

• Agent definition: A structured way to define what an agent can do, what tools it has access to, and what its goals are.
• Planning and reasoning: Built-in patterns for how agents break tasks into steps and decide what to do next.
• Tool integration: Standard interfaces for connecting agents to external tools, APIs, databases, and services.
• Memory and state: Systems for maintaining context across multiple steps of a task.
• Multi-agent coordination: Patterns for multiple agents to collaborate, delegate, and share information.
• Error handling: Strategies for recovering when a step fails or produces unexpected results.

The current landscape

Several frameworks have gained traction, each with a different philosophy:

• CrewAI: Uses a crew metaphor — agents have roles, backstories, and goals. They collaborate on tasks like a team of specialists. Emphasises simplicity and rapid prototyping.
• LangGraph: From the LangChain team. Models agent workflows as directed graphs with explicit nodes, edges, and state management. Gives developers precise control over flow.
• AutoGen (Microsoft): Focuses on multi-agent conversation patterns. Agents communicate through structured dialogue.
• Semantic Kernel (Microsoft): Designed for enterprise integration with existing .NET and Python codebases. Strong on planning and function calling.

How frameworks differ from chatbots

A chatbot responds to a single message with a single response. An agent built with a framework can receive a goal ("research our three main competitors and create a comparison table"), break it into steps, search the web, read documents, analyse data, draft output, review its own work, and deliver a finished result — all autonomously.

The framework provides the infrastructure that makes this autonomy possible, reliable, and observable.

When to use a framework

Frameworks add value when you are building agents that need to coordinate multiple steps, use multiple tools, or involve multiple agents working together. They are particularly useful for:

• Repeatable business workflows (research, reporting, data processing)
• Systems where you need visibility into what the agent is doing and why
• Multi-agent applications where coordination logic would be complex to write from scratch

Considerations for choosing

When evaluating frameworks, consider: How much control do you need over the agent's decision-making? Does the framework support the programming language your team uses? How active is the community and documentation? Does it integrate with the LLMs and tools you already use? And critically, does it provide adequate observability — can you see what the agent did, why, and where it went wrong?