An open-ended coding assistant in which multiple agents collaborate: a planner develops the solution plan, a coder writes the code, and a reviewer evaluates, recommends changes, and helps debug.
A multi-agent content system with separate researcher, writer, and editor roles. The researcher gathers web material, the writer produces an article, and the editor revises it before publication to a social media website.
AutoGen is an open-source framework from Microsoft for building and orchestrating agent-based workflows that use large language models. It provides components for defining agents, managing conversations and interactions, and integrating models and external tools.
BabyAGI is an experimental framework for building a self-building autonomous agent. Its core, the functionz framework, stores, manages, and executes functions from a database using a graph structure that tracks imports, dependent functions, and authentication secrets; it also supports automatic function loading and logging. A dashboard provides function management, updates, execution, and log viewing. The repository can be installed as a Python package and is intended for experimentation rather than production use. The original repository was archived in September 2024 and moved to a separate archive repository.
ChatDev is an open-source multi-agent framework from OpenBMB. Its original ChatDev 1.0 design models a virtual software company in which role-based agents such as a CEO, CTO, and programmer collaborate through specialized functional seminars to carry out software development tasks including design, coding, testing, and documentation. The current ChatDev 2.0, called DevAll, is a zero-code multi-agent orchestration platform. Users configure agents, workflows, and tasks to build and run customized multi-agent systems for scenarios such as software development, data visualization, 3D generation, and deep research.
CrewAI is an open-source Python framework for building multi-agent workflows. Its Crews use autonomous, role-based agents that collaborate on tasks, while its Flows provide event-driven workflow control combining precise orchestration, individual LLM calls, and Crews. The project also offers CrewAI AMP Suite, a commercial control plane for managed deployment, observability, governance, security, enterprise support, and cloud or on-premise operation. The repository provides high-level abstractions and lower-level APIs, along with documentation, examples, installation guidance, and integrations for connecting agents to language models.
Flowise is a visual, low-code platform for building AI agents and agentic workflows by connecting models, components, and third-party integration nodes through a graphical interface. Its monorepo contains a Node.js backend for API logic, a React frontend, component integrations, and autogenerated Swagger API documentation. It can be installed with npm, run with Docker, deployed self-hosted, or used through Flowise Cloud. The repository is archived and its source code is released under the Apache License 2.0.
LangChain is an open-source framework and agent engineering platform for building agents and applications powered by large language models. It chains interoperable components and third-party integrations, providing standard interfaces for models, embeddings, vector stores, retrievers, tools, and other data sources. It supports sequential agentic workflows, context and tool-call orchestration, model substitution, and application development primarily through Python; the project also provides a separate JavaScript/TypeScript library. LangChain can be used standalone or with related tools for agent orchestration, evaluation, observability, debugging, and deployment.
LangFlow is an open-source, low-code visual builder for creating and running language-model pipelines, agent workflows, and retrieval-augmented generation (RAG) applications. The project is developed and maintained by the open-source community led by the GitHub user logaretm and integrates with language-model providers and tools such as LangChain.
An agent-orchestration framework for Python that models complex, multilayered AI workflows as directed agent graphs. It manages context and tool calls, provides state checkpoints and persistence, and supports human-in-the-loop approval gates for production deployments. The framework is reported to allow existing agents to be exported/imported into IBM watsonx Orchestrate.
LlamaIndex is an open-source data framework and developer platform (originally GPT-Index) that provides connectors, indexing structures, and retrieval utilities to connect documents and external data to large language models. It is maintained by the LlamaIndex team and is used to build LLM-powered pipelines, agents, and document workflows.
Semantic Kernel is an open-source SDK developed by Microsoft for building AI applications that combine large language models with conventional code, memory, and external connectors. It provides orchestration primitives, memory abstractions, tooling for prompts and skills, and SDKs for .NET and Python.
Searchable transcript of Agentic AI Frameworks Explained: Workflows, Multi-Agent, & Production — IBM Technology (11:54). Search for a phrase, then click its timestamp to jump straight to that moment in the video.
Captions sourced from the original video on YouTube, published by IBM Technology. The video, its captions and all related intellectual property remain the property of their respective owners; AINotes claims no ownership. Provided for research, accessibility and search — see the Transcript Notice and Copyright Policy.
00:00 Hi, let me guess. The world around you is abuzz with agentic AI systems and their massive potential. So you decide to go off and build an agentic system. You look for the best available framework out there. And now, all of a sudden, you have 17 GitHub tabs open, five medium blocks bookmarked, and you are still clueless on how to proceed. Yes, we've all been there.
00:24 LangChain, LangGraph, Crew AI, AutoGen, Semantic Kernel. There are so many powerful frameworks out there, but which one would you pick? And in order to answer that question, you first need to understand which type of agentic AI system you want to build. In this video, we are going to cover five types of most common agent AI projects or systems. We are going discuss an example for each of those and also list frameworks which are best suited for those types of agentic AI systems and projects.
00:57 First, let's get some basics out of the way. What exactly is an agentic AI framework? An agentic framework is a toolkit for building agentic AI systems. Let's understand with an example. Let's say you have an agent to analyze some sales data. This agent goes to a database and extracts the relevant data. It analyzes the data, maybe runs some calculations, and then generates a report and crafts a response that can be sent to the concerned person.
01:24 Now, there's a lot going on here, and there's lot of coordination that's needed. And it gets even more complex when you have multiple agents working. And that's exactly why you need an agentic AI framework. The framework is like a building block. Unlike a chatbot application where you just ask a question and get an answer, the agentic system actually does a lot of planning, acting, and iterating.
01:53 It is because of this complexity involved that we need an agentic AI framework. These are like building blocks that help us deploy and manage agentic AI systems. Now these have some predefined functions that help us build agentic AI systems with more ease and agility, such as we have predefined architectures. We might have integration and monitoring tools.
02:30 We might also have some task management capabilities. And communication protocols. Together, these features and functionalities give agentic AI frameworks the capability to allow us to build these systems with ease. Like we discussed before, there are multiple agentic frameworks available out there, but they're not all competing to do the same type of tasks.
02:58 In fact, they're optimized for different types of agentic AI systems. Most agentic AI systems and projects we are working on today fall into one of the five categories. First, we have linear workflows. We have autonomous AI agents or autonomous multi-agent systems. We have role-based AI systems, production orchestration systems, and then we have rapid prototyping.
03:21 Let's dive in and understand each of these in more detail. Let's start with the simplest one, linear workflows. Now, this type of an agentic AI system is where things progress in a step-by-step fashion. It is more predictable what's gonna happen next. And the steps follow a certain sequence. For example, consider a customer-facing application, let's say a customer support agent.
03:53 The role of this agent is when a user asks a question, the agent is going to take the question and search the knowledge base for relevant responses. It's then going to craft a response and send it back to the user and maybe take an additional action such as creating a support ticket. Now, if you observe, these steps are progressing in a certain fashion, in a certainly sequence, and these systems are more useful when you need the flows to be more reliable.
04:22 There isn't a need for multiple agents to collaborate to make this happen. And that gives you more control on how things progress. A good example of frameworks that are suitable for this kind of a setup include LangChain. And LlamaIndex. LangChain is more suited for setups where multiple steps need to be happening in a certain sequence. LlamaIndex is highly suitable for heavy applications that are heavy on the data retrieval and indexing.
05:03 For more complex setups, you could also use LangGraph, which is also by LangChain. Next, we have the autonomous agentic AI systems. In these systems, you typically give AI a goal. And have it figure out how to accomplish it. So in this system, it's very common to see multiple agents collaborating together. These agents talk to each other to accomplish the common goal.
05:40 A good example of this could be an AI coding assistant. You could have a planner agent that plans the solution for you. You could a coder agent that actually writes the code for you, and a reviewer agent, that is reviewing the code, making recommendations, and also helping with the debugging. These agents are constantly talking to each other in order to give the best code possible to you.
06:02 So, in this kind of setup, the problem is usually open ended. And that's the kind of problems this kind of setup is most helpful for. So, frameworks that work best for this kind of a scenario include AutoGen. You could also use experimental setups like Baby AGI. And CrewAI could also be helpful for designing these kind of systems where problems are open-ended and multiple agents are collaborating together to achieve a shared goal.
06:43 Next, we have the role-based agentic AI systems. These are kind of similar to the autonomous agentic systems where there are multiple agents collaborating. So it is also a multi-agent setup. But what makes it different is that each agent within the setup has a defined role. They are still communicating with each other to accomplish that goal, but they are operating within the confines or the constraints posed by their role descriptions.
07:20 They are working together, but with clear boundaries. A good example of this could be a content generation agent. Here, you could have a researcher agent that goes on the web and fetches all the material that's needed to write a piece of content. There could be writer agent that looks at all the content that has been fetched and writes up an article.
07:39 That goes out on a social media website, let's say. And then there could be an editor agent that's looking at the article that has been written and make some edits to it. Now, they have very clearly defined roles and they don't go into other agents roles when they do this. They have discussions, but they're strictly confined to the description that has given to them for their particular roles.
08:01 A good framework that is applicable here is CrewAI. But you could also use AutoGen with some structures around it for this kind of an agentic AI systems. There are also some niche frameworks that are applicable to very specific tasks. Like, for example, for software development kind of tasks, you have ChatDev. So these are the kind of frameworks that you would use for role-based agentic AI systems.
08:34 Next up, we have the production orchestration systems. Like the name suggests, this is when AI moves out of the experimentation phase and gets real or moves into a real-world system. These kind of systems require deep integration with APIs, databases, and business workflows. Consider the example of an AI operations agent. This agent detects alerts, searches the documentation for the alerts, and then runs some automation scripts and sends summaries in a real world scenario within an organization.
09:22 Good examples of frameworks that are suitable for this kind of an AI system, include agent framework, which essentially is a combination of semantic kernel and autogen. Another good example here is LangGraph. Which works for well-structured, multilayered applications. Ancient framework is suitable both for orchestration as well as for running autonomous workflows.
09:57 Last but not least, you have the rapid prototyping. So you always don't need a perfect architecture. You just need to check if your idea would work or not. These types of systems are best when you need to quickly validate ideas. It helps you build quick prototypes. To see if you can bring your ideas to reality. These kind of systems are where you ideally have a user interface where you can drag components and bring them onto a canvas to build quick workflows and test ideas.
10:36 Examples of tools or frameworks that are useful here include LangFlow. And Flowise. These tools offer you a good graphical user interface where you can bring components onto a canvas and connect models and workflows and quickly test out your ideas that you can later on take into production. These are very quick for rapid prototyping and hence the name.
11:05 So when choosing a framework, do not ask which framework is the best. Instead ask what kind of system am I trying to build? If it's predictable, use a workflow approach. If it is exploratory, use the autonomous agents. If it needs teamwork, use role-based systems. If it's going into production, use the production orchestration frameworks. And if you are testing ideas, use the rapid prototyping tools.
11:28 Because the right framework depends on whether you're building a pipeline, a team of agents, or a production AI system. Which agentic AI framework is your typical go-to? Feel free to comment below and don't forget to like and subscribe. Thank you.