Agentic AIs
My recent months in research and development of AI agent systems.
The field of AI is rapidly growing with AI Agents to drive most of that technological wave. AI Agents are autonomous system that are able to make independent decision based on real world inputs. Agentic AI space is rapidly growing with the market projected to soar from $5.1 billion in 2024 to $47.1 billion by 2030 [1]. This article explores agentic AI trends, use cases and some tools that I have discovered through the journey of building them.
Use cases
There are many use cases of agentic AI and most of them cover e-commerce, finance, marketing, human resource fields, healthcare etc.
Web3 industry, social media platforms like Twitter, Discord, and Farcaster are essential for connecting with users, obtaining cutting-edge information, and making trading decisions. As an increasing number of Key Opinion Leaders (KOLs) flock to these platforms, the information they disseminate becomes more complex and fragmented. Navigating this landscape to acquire organic insights and critically assess the credibility of KOLs is a universal challenge for traders. An exemplary Agent would enable users to sift through the vast information pool, distilling valuable intelligence without succumbing to information overload, and serving as a genuine intermediary in social media interactions with other users or agents. [2] Spending a lot of time on X platform I discovered agents like AIXBT and MindAIAGENT that do successfully aggregate information from blockchains and provide useful insight and time sensitive events.
I find this use case for entertainment quite interesting as well as it has never been explored before Andy Ayrey has launched the Truth Terminal. Probably one of the first times where two Claude AI instances were left alone to talk to each other and conversations where saved in .txt files which were later used in building an AI character on X platform.
Quite an interesting space to explore is e-commerce with tremendous potential. Agent Kitsune, an animated fox that can order you coffee through it’s terminal. It has full Shopify inventory integration and accepts stable coin payments.
Builder frameworks
With the birth of truth terminal we saw new developer frameworks emerging like ElizaOS, LangChain, AgentKit and many more. What kind of systems can you build with these tools? Let’s have close look at them.
Eliza, a pioneering open-source web3-friendly agentic operating system designed to bridge this gap. Eliza is the first of its kind, offering a platform that makes the deployment of web3 applications not only possible but also effortless. We emphasize that every aspect of Eliza is crafted as a regular Typescript program, ensuring that it remains under the full control of its users while also providing seamless integration with web3 functionalities. This includes, but is not limited to, reading and writing blockchain data, interacting with smart contracts, and much more functionality. Connecting Agents with outside world systems has proven obvious traction in the developer community.
LangChain is an open-source framework designed to simplify the development of applications using large language models (LLMs). It provides tools for connecting LLMs to other sources of data and computation, allowing developers to build context-aware, reasoning applications. LangChain offers components for prompt management, memory systems, agent construction, and tool integration, making it easier to create sophisticated AI assistants and automation systems. What I like about it that it has pdf loaders so one can build Agent knowledge on already existing hard to search information sources.
AgentKit is a modern toolkit for building AI agents with advanced reasoning capabilities. It provides developers with infrastructure for creating, testing, and deploying AI agents that can understand complex instructions, use tools effectively, and perform multi-step tasks. AgentKit focuses on agent reliability, offering features for structured reasoning, tool use orchestration, and improved error handling to help developers create more dependable AI assistants.
Character
Character development in ElizaOS has been quite a feature where a user can focus on a lot of personality traits. This feature can be configured for crafting an AI character that mimics the communication style of famous people or brand new personalities itself. There are several key configuration elements:
Basic Settings: Including the name, model provider (OpenAI, Anthropic), and voice settings (Eleven Labs).
Personality Components:
"bio" entries that define key talking points
"lore" that establishes background knowledge and narratives
"knowledge" that specifies what the character claims to know
"topics" of focus for the character
Communication Style: Detailed styling guidelines for different contexts (all communication, chat, and posts), including use of capitalization, specific phrases, and rhetorical techniques.
Message Examples: Sample conversations showing how the character responds to specific questions, providing training examples for the AI.
Post Examples: Sample standalone messages the character might produce.
You could use similar configuration files to create other AI characters by:
Creating a news anchor character with formal language, neutral stance, and knowledge of current events
Building a customer service agent with empathetic language, product knowledge, and problem-solving abilities
Designing a creative writing assistant with literary knowledge, varied vocabulary, and different writing styles
Developing an educational character for children with simple language, encouraging tone, and age-appropriate knowledge
These configurations help define not just what an AI agent knows, but how it communicates that knowledge through specific language patterns, personality traits, and communication styles.
Building personality of an agent is amazing, but it gets even more interesting when you connect it to real world integrations. I’m talking data extraction, payments, purchasing through your product APIs and even meme generation. It also can carry out media processings: PDF, URLs, Audio transcription, Video processing, Image analysis, Conversation summarizations. Building on ElizaOS is effortless when it comes to third service provider integrations like web-search with Tavily, video generation APIs like Luma Labs or meme generation API with ImgFlip. There’s endless possibilities to what kind of service you’d like to connect your agent that you’re building.
Furthermore, I’d like to talk about the difference in between APIs and Operators. We all know what an API is, right? In case you don’t its an Application Programming Interface (API) is a software intermediary that allows different applications, systems, or services to communicate with each other. To be short, companies do that so others could build services on top of them. Like X platform allowing tweeting programmatically. If you are working in tech, you probably realized that not all services we interact with have API access or its heavily limited. Yet its still one of the best ways to build integrations.
On the other hand, a new way of interacting and building actions around the web is operators. OpenAI's Operator is a specific AI agent that can navigate the web like a human, automating tasks such as booking flights or ordering groceries. It uses a Computer-Using Agent (CUA) model to interact with websites through virtual mouse and keyboard inputs, allowing it to perform tasks independently. It seems like this way of data extraction can go beyond traditional API services for those companies that have no interface. Here’s an example of me running a prompt on a custom built operator service using playwright to build a meme with captions and share the url. As a programmer I don’t have to pay 200$ for chatgpt operator subscription ^^:
I’m playing here with operator and LLM using playwright and Chromium browser
OpenAI Operator and APIs offer different approaches for accessing AI capabilities. The OpenAI Operator provides a conversational interface where users can interact with AI models through natural language commands, making it more accessible for non-technical users and enabling dynamic, iterative workflows. APIs, on the other hand, require programming knowledge but offer more customization, integration possibilities, and consistent performance at scale. APIs allow developers to embed AI functionality directly into applications with precise control over inputs and outputs, while the Operator prioritizes ease of use and flexibility. The choice between them ultimately depends on use case requirements, technical expertise, and whether the priority is rapid exploration or production deployment.
The agentic AI landscape is evolving rapidly and these autonomous systems span diverse industries including Web3, e-commerce, entertainment, and social media. Developer frameworks like ElizaOS, LangChain, and AgentKit are enabling the creation of increasingly sophisticated agents with distinct personalities and capabilities. The distinction between traditional APIs and newer operator-based approaches represents different pathways for integration—APIs offering programmable control and customization, while operators enable more human-like web navigation for tasks without API access. As this technology continues to mature, we're seeing the emergence of AI agents that can not only process information but also interact meaningfully with real-world systems, creating new possibilities for automation and user engagement.
Citations
1. AI agents: 2025 predictions | MarTech, accessed January 11, 2025, https://martech.org/ai-agents-2025-predictions/
2. ElizaOS white-paper https://arxiv.org/pdf/2501.06781