The AI landscape is evolving at an unprecedented pace, making it increasingly challenging to categorize and understand new developments. To simplify this complexity, I’ve structured AI into three distinct layers. Each layer represents a different level of capability and autonomy, providing a clear framework to analyze AI tools, applications, and trends.
Why Classification Matters
Most AI definitions are shaped by influential companies or individuals, and these definitions gain widespread adoption. For instance, terms like “actors” are used by Apify, while organizations like OpenAI and Anthropic influence AI terminology.
Apify is the largest ecosystem where developers build, deploy, and publish web scrapers, AI agents, and automation tools. We call them Actors.
A structured approach, similar to how blockchain technology is classified into layers, can bring clarity to the AI ecosystem. But before layers we can also largely divide AI’s into Implicit and Explicit AI’s.
Implicit AI
Implicit AI refers to artificial intelligence that’s quietly embedded into everyday products, services, or platforms. Unlike Explicit AI, which you interact with directly (e.g. ChatGPT or Midjourney), Implicit AI works behind the scenes to enhance user experience, automate decisions, or optimize systems without drawing attention to itself.
Explicit AI
Explicit AI refers to artificial intelligence systems that users engage with directly and intentionally. These are the AI tools, agents, or platforms where the AI is front and center—clearly labeled, visible, and often the core feature of the product or service. You know you’re using AI when you interact with these tools.
The Four Layers of AI
Layer 0: AI Frameworks (TensorFlow, PyTorch, HuggingFace, JAX)
This is the foundational layer where the building blocks of AI models live. These frameworks provide the infrastructure for building, training, and deploying models.
Mobile ML, edge devices, applied industry production
JAX
Google’s internal large model research (e.g., Gemini, PaLM, Imagen)
🧪 Why PyTorch Is Dominating
Dynamic computation graph = Easier to debug and experiment with.
Pythonic syntax = Feels native to developers.
Huge ecosystem = HuggingFace, open-source LLMs, computer vision, etc.
☁️ What About TensorFlow?
More common in applied settings, less so in cutting-edge research.
Still widely used in production (mobile, embedded, Google tools).
TensorFlow Lite, TF.js, and TensorFlow Serving are production deployment tools.
🔹 Layer 1 – Pre-trained Models
At the foundation of AI are the Core Models—large-scale AI engines trained on vast datasets. These models generate raw outputs such as text, images, audio, or video. They serve as the building blocks for more complex AI applications but do not inherently possess workflow automation or task execution capabilities.
Example solutions are :
Agents
RAG
tool use
LLM’s with memory
multi-agent systems
Fine-tuning a model/transformer
Generative Adversarial Network. (GANs)
Prompt-tuned wrapper
Examples:
GPT-4 (OpenAI)
Claude (Anthropic)
Stable Diffusion (for image generation)
🔹 Layer 2 – AI Agents, Applications & Workflows
This layer consists of tools and platforms that utilize Layer 1 models but enhance them with user interfaces, workflow automation, and domain-specific optimizations. These AI-powered tools integrate core models into practical use cases but do not exhibit autonomous or agentic behavior. RAG based implementations would also fall in this classification.
AI assistants, advisors, and agents have different capabilities and use cases, according to IDC. Credit:IDC
🔹 Layer 3 – Agentic AI & Autonomous Systems
At the top of the AI hierarchy are Agentic AI systems, which go beyond simple inputs and outputs. These systems can plan, execute tasks, and iteratively improve with minimal human intervention. They often function as AI agents capable of self-directed learning, making them a major leap toward autonomous AI.
Specific definition from CIO Agentic AI is a type of artificial intelligence that adapts to new situations, learns from experiences, and operates independently to pursue goals without human intervention . This is a significant departure from traditional AI, which typically follows preset rules or algorithms . Unlike traditional AI, agentic AI empowers systems to act autonomously, making decisions and executing tasks with little to no human involvement . Furthermore, agentic AI emphasizes proactive problem-solving and complex task execution, unlike generative AI tools that focus on content creation . This ability to autonomously plan, execute, and adapt their actions to achieve complex goals sets agentic AI apart from traditional automation, which follows predefined rules .
Understanding Agentic AI
The term “Agentic AI” has gained popularity recently, but there is no single individual or organization explicitly credited with coining or popularizing it. It has emerged as a key concept in discussions of advanced AI systems, often highlighted by technology experts and companies such as IBM, Forbes, and KNIME.
Definition from computerworld.com
Agentic AI is a type of artificial intelligence that adapts to new situations, learns from experiences, and operates independently to pursue goals without human intervention . This is a significant departure from traditional AI, which typically follows preset rules or algorithms . Unlike traditional AI, agentic AI empowers systems to act autonomously, making decisions and executing tasks with little to no human involvement. Furthermore, agentic AI emphasizes proactive problem-solving and complex task execution, unlike generative AI tools that focus on content creation . This ability to autonomously plan, execute, and adapt their actions to achieve complex goals sets agentic AI apart from traditional automation, which follows predefined rules.
General Meaning of Agentic AI
Agentic AI refers to artificial intelligence systems that exhibit autonomy, decision-making capabilities, and adaptability. These systems are designed to act as independent agents, capable of pursuing complex goals, making decisions, and interacting with their environments with minimal or no human oversight. Key characteristics include:
Autonomy: The ability to act independently without constant human input.
Goal-Oriented Behavior: Pursuing specific objectives by analyzing data and planning actions.
Adaptability: Adjusting strategies and decisions in response to changing conditions.
Proactive Execution: Taking initiative to optimize workflows or solve problems.
Agentic AI differs from traditional AI, which typically operates within predefined rules or tasks, by emphasizing independent reasoning and adaptive behavior. Applications range from robotics and autonomous vehicles to intelligent assistants and enterprise automation.
Examples of Agentic AI in Various Industries
1. E-Commerce
Amazon’s Recommendation System: Uses agentic AI to autonomously provide personalized product recommendations, contributing to 35% of the company’s revenue.
Dynamic Pricing Systems: Platforms like Uber use AI agents to adjust prices in real-time based on demand, competition, and other factors.
2. Autonomous Vehicles
Waymo’s Self-Driving Cars: Employ multiple agentic AI systems for navigation, decision-making, and real-time responses, operational in cities like Phoenix and San Francisco.
3. Customer Support
Chatbase AI Agents: Handle complex support tasks such as processing refunds, resetting passwords, and providing product recommendations autonomously, reducing support tickets by 65% for clients.
4. Healthcare
Google’s Health AI: Diagnoses medical conditions and analyzes medical images with high accuracy, such as surpassing dermatologists in diagnosing skin cancer.
5. Manufacturing
AI-Powered Robots: Perform tasks like welding, painting, and assembly autonomously, optimizing production time while maintaining quality standards.
6. Software Development
GT Edge AI: Automates code conversion (e.g., COBOL to Java) by retrieving source code, processing it via LLMs, and updating repositories autonomously.
GitHub Copilot: Provides real-time code suggestions and auto-completions to streamline software development.
7. Content Recommendation
Netflix and Spotify: Use learning agents to provide personalized content recommendations based on user preferences and behavior.
8. HR and Enterprise Management
Agentic AI is used for tasks like recruitment automation, onboarding processes, performance management, and compliance handling in HR workflows.
The Future of AI Categorization
As AI continues to advance, understanding where a tool or model fits within this layered framework will become increasingly important. By using clear classifications, we can navigate the evolving AI landscape with greater clarity and precision.
Stay informed and ahead in the AI revolution by analyzing developments through this structured lens.
