Always know some basics - Notes

16 Aug, 2025

The book Principles of Building AI Agents (2nd Edition) by Sam Bhagwat is short and explains concepts in really straightforward ways.

Here are some simple notes from part 1 that I’ve found helpful

• AI apps today don’t require building a model from scratch—you just need to know how to use LLMs effectively.

• Big vs Small Models

Big models → more accurate, but slower & costly

Small models → faster & cheaper, but less accurate

• Context Window

Think of it as the model’s memory.

Bigger memory → can handle long chats or documents.

Smaller memory → forgets things quicker.

• Reasoning Models vs Chat Models

Reasoning Models: Designed for complex problem-solving. They break tasks into steps, check their work, and aim for accuracy. Best for logic puzzles, code debugging, planning, or analysis.

Chat Models: Optimised for smooth conversations. Great at summaries, drafting text, answering FAQs, and quick back-and-forth.

-- Key difference: Chat models focus on being helpful and fast. Reasoning models focus on being accurate and thoughtful (but take longer and cost more).

• Prompting Styles

Zero-Shot → Just ask the question to the LLM.

One-Shot → Show one example to the LLM.

Few-Shot → Show a few examples for consistency.

• System Prompts

Tell the AI how to behave.

Define personality, tone, or rules for the model. Example: 'You are a helpful tutor who explains concepts simply.'

-- Useful for controlling style and behaviour.

The trick isn’t in building a new AI—it’s in choosing the right model (chat vs reasoning), giving it the right context, and asking the right way.