How to Build an AI Agent with Claude Code (Claude AI Agent Tutorial)
This tutorial explains how to build AI agents using Claude's desktop code workspace without coding, focusing on three distinct levels of AI work (basic chat, builder mode, and agentic work) and providing practical workflows for research and content repurposing tasks.
Summary
The transcript outlines a comprehensive guide to understanding and building AI agents with Claude. The speaker establishes three hierarchical levels of AI interaction: Level 1 (basic chat/search engine functionality), Level 2 (builder mode where users create specific outputs but manage each step), and Level 3 (agentic work where users set goals and the AI independently executes multi-step processes). Real agents are distinguished from sophisticated chatbots by three characteristics: following a defined process with multiple stages rather than single-response answers, making decisions under uncertainty by asking clarifying questions instead of guessing, and requiring clarification before execution to avoid misunderstandings. The core technical setup involves three components: a workflow file (a written SOP in plain English), the agent itself (Claude acting as coordinator), and a toolset (built-in capabilities like file creation and text analysis). The critical foundation is a claude.md file that provides persistent project context including project description, personal background, working rules, and folder structure—this single file is presented as more valuable than most prompt engineering techniques. The speaker emphasizes planning mode, where agents must present their approach before execution, catching problematic assumptions early. Two practical workflows are detailed: a research agent that asks clarifying questions, forms a plan, researches thoroughly, and saves structured reports; and a repurposing agent that takes a single long-form script and generates three 60-second shorts plus a social media pack with PDFs. Common mistakes are identified: skipping claude.md, using vague goals, skipping plan review, not requiring clarifying questions, and trying to build too many workflows simultaneously. The recommended progression starts with simple self-contained tasks, moves to iterative workflows with refinement rounds, and eventually scales to multi-workflow systems where outputs from one inform another.
Key Insights
- The speaker argues that the intelligence in effective agent systems lies in the instructions and workflow design rather than the technology stack or tool complexity, stating that a thoughtfully designed workflow with basic tools will outperform a poorly designed workflow with sophisticated integrations.
- The speaker identifies that most poor AI output results from the model misunderstanding what the user actually wanted rather than technical limitations, and positions clarification before execution as the step where most quality lives in agent workflows.
- The speaker explains that a single claude.md file written once and read in every session provides more value than most prompt engineering tricks found online, functioning as a persistent onboarding document that eliminates the need to reexplain preferences and rules.
- The speaker argues that planning mode—requiring agents to show their plan before execution—saves significant time by catching incorrect assumptions during the 2-minute review phase rather than requiring 10 minutes of cleanup after the agent has already committed to a wrong direction.
- The speaker demonstrates that persistent project context compounds over time, allowing refinements and additions to build on previous work without starting from zero, fundamentally changing the relationship with AI from a copy-paste loop to iterative development of a living body of work.
Topics
Transcript
[0:00] Most people are still using Claude like a fancy search engine. Ask a question, copy the answer, ask the next one. There's a way smarter way to work with it. And once you see it, you can't go back. Let me show you what an AI agent actually is and how to build one inside Claude without writing a single line of code. Before we build anything, I want to get clear on one thing because the term AI agent gets thrown around constantly and most explanations are either too technical or too vague to be [0:30] useful. Here's how I think about it. There are three distinct levels of working with AI and most people are stuck at…
Full transcript available for MurmurCast members
Sign Up to AccessMore from AI Master
30 Claude Features You Need to Know NOW (Become a PRO!)
A tutorial video covering 30 advanced Claude AI features, including adaptive thinking triggers, prompt engineering frameworks, memory and style settings, connectors, research mode, skills, Claude Code, and Co-work. The presenter explains how to configure Claude's environment for maximum productivity across writing, coding, and data workflows. Several features are described as significantly improving output quality when properly activated.
How to Create Scalable Business App With AI (No Code, No Wasted Budget)
The video demonstrates building a full event management app in Softr using a hybrid AI + visual editor approach, arguing that AI should only handle initial scaffolding while visual tools handle permissions, workflows, and logic. The creator compares Softr against Lovable and Replit, showing how the latter two generate first and ask questions never, leading to costly re-prompts. The core thesis is that the 'last 20%' of any AI-built app is where budgets quietly drain, and visual editors solve that problem.
How to Make an AI Film With the Same Character in Every Scene
A creator demonstrates how to build a consistent, 60-second AI short film using Imagine Art's Film Studio, focusing on the importance of pre-production planning, shot lists, and storyboard mode for character consistency. The walkthrough covers the full pipeline from story concept through generation, editing, and export within a single platform. The video is sponsored by Imagine Art.
NEW Gemini Features Explained — How to Use Google’s Latest AI Upgrade
The video reviews Google's Gemini 3.5 Flash model across multiple use cases including multimodal vision, video analysis, document processing, and agentic workflows. The presenter argues Flash delivers Pro-tier performance at a significantly lower price point than competitors like Claude and GPT. However, the review also surfaces three under-reported weaknesses: degraded long-context retrieval, increased verbosity, and a silent default thinking level downgrade.
Master Nano Banana 2 Instantly – The SECRET JSON Prompt Engineering Formula
This tutorial explains how to use JSON-structured prompts with Nano Banana 2 (an AI image generation model) to achieve precise, reproducible control over individual image elements. The presenter demonstrates how separating prompt elements into labeled JSON fields prevents unwanted changes during image editing. The video covers use cases including style transfer, character consistency, lighting control, object swapping, camera perspective transfer, and video generation.