An interactive explainer
A coding agent looks like one program working on your repo. It is not. The model can only produce text; a plain program on your machine reads that text, performs the one action it names, and pastes the result back. And it calls the model fresh every step — so the agent's whole memory is a conversation that gets re-sent, in full, every single time.
AI agents · Tool calling · The agent loop
Developer Tools · Intermediate · 12 min · Why a coding agent forgets everything between steps — an interactive explainer on ViLearn, with a simulation and a knowledge check. All explainers.
Related: Why sending the model more can make the answer worse · How an LLM picks the next word · How Git actually stores your work