Layer 4
Applications
How a model becomes something that is actually of use to somebody.
A language model on its own is not a product. The application layer is where the benefit is decided: who may see what, where the correct information is found, when a person has to approve the result, and who answers for a mistake.
What happens in this layer
01
The prompt says what is being done right now
Instructions and context: the user's question, the application's rules, examples, documents, and a description of the tools available.
02
Retrieval brings in the facts
Before answering, the system searches the company's own documents and hands the passages it found to the model. The model itself does not change, but the sources can be shown to the reader.
03
Fine-tuning changes how it behaves
The model's numbers are adjusted for a purpose: a house style, a classification task, the conventions of a particular field. A poor way to add fresh facts.
04
An agent is a loop
The model sizes up the task, picks a tool, looks at the result, corrects course and continues. The tools, the permissions and the checks decide the quality, not the eloquence.
A common misconception
Key terms
Whole glossary →Articles on this topic
Vibe coding and AI-assisted programming
Two different ways of working get grouped under one label, though the distinction that matters is whether the generated code is reviewed. What each approach is good for, what remains after the demo, and what productivity research actually shows.
AI use cases in plain language
Thirty concrete use cases, from the home to the forest, from the production line to a sports club. What the machine does in each, where it goes quietly wrong, and what happens to a use case after it is chosen.
Agents, orchestration and use cases
What a multi-agent system actually is, the four structures you find in one, what each is good for, and when a single agent is enough.
When an agent believes the wrong text
Prompt injection has topped the OWASP list for a third year, and there is no fix. Why not, and what to do instead.
Prompt, retrieval or fine-tuning
Three ways to make a model do what you want, and when each one is the right tool.
What an agent actually is
An agent is not magic. It is a loop: the machine tries something, looks at the result and corrects.