A Large Language Model (LLM) is an artificial intelligence system trained on enormous amounts of text data. Through training, the model learns patterns in language and can generate human-like responses, summarize information, answer questions, write code, and assist with a wide variety of tasks.
Examples include:
Although these systems often appear intelligent, they are fundamentally prediction engines. Given a prompt, they generate the most likely next sequence of words based on what they learned during training.
One of the first concepts that confuses newcomers is the difference between a model and a framework.
An AI model is the intelligence itself.
A model is trained on large amounts of data and learns patterns that allow it to:
However, a model by itself has significant limitations:
Once trained, the model’s core knowledge is essentially frozen until it is retrained.
An AI framework is the software infrastructure used to build applications around a model.
Frameworks help developers:
Popular examples include:
A useful analogy is to think of the model as the engine and the framework as the rest of the vehicle.
The engine generates power, but without a chassis, steering wheel, dashboard, and wheels, it cannot take you anywhere. Similarly, a model provides intelligence, while a framework turns that intelligence into a usable application.
An LLM harness is the operational environment that surrounds and manages a running AI model.
If the model is the brain, the harness is the body.
A harness provides the tools, memory management, safety controls, and execution environment needed for an AI system to interact with the outside world.
Without a harness, an LLM is simply a sophisticated text generator.
With a harness, the model can:
Modern AI assistants often perform tasks that require multiple steps.
For example:
A raw LLM cannot perform this workflow because it cannot directly interact with a computer.
The harness provides the missing capabilities by:
This process is commonly called an agentic loop.
Although the terms are sometimes used interchangeably, they serve different purposes.
| AI Framework | LLM Harness |
|---|---|
| Development toolkit | Runtime environment |
| Used while building applications | Used while running applications |
| Provides reusable libraries and APIs | Provides execution, memory, and control |
| Standardizes development | Enables autonomous operation |
| Examples: LangChain, LlamaIndex | Examples: Claude Code, Ollama, Codex |
In many cases, developers use a framework to build a harness.
Another common source of confusion is the difference between an AI model and the products built around it.
Claude is the family name for Anthropic’s language models.
Within that family are several model tiers:
Many people assume that larger models simply have larger context windows. In reality, recent Claude models often share similar context window sizes.
The primary difference is model capacity.
A larger model contains a larger neural network with more parameters, allowing it to reason more effectively and understand subtle relationships in information.
Claude Code is an excellent example of an LLM harness.
The underlying model provides intelligence, but Claude Code provides:
When the model determines that a command should be executed, the harness performs the action, captures the results, and feeds them back to the model.
This allows the system to function as an autonomous coding assistant rather than a simple chatbot.
Ollama is another example of an LLM harness.
Its purpose is to simplify running AI models on local hardware.
Ollama handles:
Unlike development frameworks, Ollama focuses on operating and serving models rather than helping developers build application logic.
This makes it an ideal starting point for anyone interested in experimenting with local AI systems.
This article is the first in a growing collection of practical AI notes, tutorials, and observations. Future topics will include:
The AI landscape changes rapidly, and part of the goal of this section is to document lessons learned along the way.
