AI Agents, Clearly Explained
About this video
Jeff Su builds a plain-English ladder from chatbots to agents in three levels. Level one is the large language model itself: it generates text from training data but knows nothing private, such as your calendar, and it stays passive until prompted. Level two is an AI workflow, where a human sets a predefined path for the model to follow, for example checking Google Calendar or a weather API before answering; retrieval-augmented generation, or RAG, is explained simply as looking things up before answering, and he shows a Make.com workflow that collects news links in Sheets, summarises them with Perplexity and drafts social posts with Claude on a daily schedule. Level three is an AI agent, where the decisive change is that the LLM replaces the human as decision-maker: it reasons about the best approach, acts through tools, and iterates by critiquing its own output, a pattern he identifies as the ReAct framework. He closes with a video-search agent demo and a recap diagram contrasting who makes the decisions at each level.
Key takeaways
- LLMs alone are passive and know no private or company data
- A workflow follows a predefined path set by a human; RAG is just lookup before answering
- An agent exists only when the LLM itself becomes the decision-maker
- Agents reason, act through tools, and iterate on their own output — the ReAct pattern
- Adding more steps to a workflow does not make it an agent
Episode details: published April 8, 2025 · 10:09 long. Summary written by Self Growth Videos from the episode itself. Business results and any income figures mentioned are Jeff Su’s own claims, shared for education — they are not typical or guaranteed.
← Previous: How to Build a Prompts Database in Notion (template included)! · Next: A Day in My Life at Google (8 years in) →
All Jeff Su videos · Jeff Su profile · Jeff Su (official site)
