Is it really just an LLM with access to external APIs?
Pretty much. Let’s clear the air by saying that when we talk about AI agents, we’re primarily referring to LLM (large language model) agents. Imagine using a general-purpose LLM like ChatGPT, but instead of relying on it “as is,” you equip it with tools to expand its capabilities.
For instance, let’s say you ask ChatGPT about the weather tomorrow. By itself, an LLM can’t answer this because it doesn’t have real-time access to data. However, if we provide it with a tool — like an API that takes a date and location as input and returns weather information — it can retrieve and present that data. This is an example of an LLM leveraging an external tool.
What’s fascinating is that these tools can be virtually any piece of code, such as a weather API, a calculator, or even a complex database query. A practical example would be ChatGPT Search, which acts as an agent. When tasked with searching the web, it uses a dedicated tool to process a query and return internet results.
What’s even more interesting is that a tool can also be another AI agent. This creates the possibility of agents calling other agents in a recursive manner, each leveraging the others’ strengths. The real challenge, then, is designing architectures that distribute tasks effectively across these agents, ensuring they work together seamlessly in agentic projects.
Where to start when building an AI Agent?
From scratch. For now, at least. While much of coding has been consolidated into libraries and reusable functions, AI agent development is still in its early days.
Some startups are popping up with marketplaces of closed-source tools that can be integrated into your projects (examples could include AssemblyAI or LangChain’s toolsets), and you want more flexibility, there is always GitHub. Platforms like mkinf are stepping in to bridge this gap by hosting customizable tools you can integrate with just an API call.
What are the frameworks for?
Frameworks like LangChain and Haystack simplify the complex orchestration required to make AI agents function effectively. Think of them as the glue that holds everything together. When you’re building an AI agent, you’re essentially creating a system that needs to juggle multiple tasks: connecting to tools, managing workflows, keeping track of conversations, and sometimes even chaining multiple agents together. That’s where frameworks step in.
The Future of AI Agents
AI agents are undoubtedly a fascinating and rapidly evolving field. They promise a level of autonomy and adaptability that could redefine how we interact with technology. From automating complex workflows to enabling new levels of collaboration between systems, the possibilities seem endless.
Whether you’re a curious developer, a business looking to integrate AI agents, or just someone intrigued by the concept, there’s no better time to dive in. The field is still young, and the opportunity to shape its future is wide open.
Don’t miss out on what’s coming and follow mkinf on X or join our slack community
Top comments (0)