How do AI agents actually use tools like web search? The model can't magically go online, it can only predict text. So when it wants to search, it outputs a structured message that looks like code, saying it wants to use the search tool. The code running the model (the agent harness) sees that, runs the tool, gets the result, and hands it back to the model to process. That handoff (model decides, harness executes) is the whole way AI agents work at scale. #techwithtim #ai #aiengineer

Do AI models actually have memory? No. On their own they store nothing, so ask a follow-up and the model forgets what you said unless it's inside the context. So how does an agent stay coherent across multiple steps? You feed the whole conversation back in. Every single turn, the entire history gets sent into the context window again. The model isn't remembering; it's rereading everything that happened, every time. #techwithtim #ai #aiengineer

How do AI agents get long-term memory or access to your documents? A vector database. The short version: you convert your data into embeddings (numerical representations of meaning), store them, then give the agent a search tool so it pulls only the relevant information for whatever you're asking. This pattern is called RAG (retrieval augmented generation), and it's how agents answer questions about documents and remember long-term facts even when the context window is full of the current conversation. #techwithtim #ai #aiengineer

This agent worked while I slept. Literally week one of what we teach inside AI Agent Builders, free. Link in the description and pinned comment.