How to Build Agentic Workflows With Claude Code

TL;DR

Agentic workflows use an AI agent to help plan, build, test, and refine multi-step automations, while deployed versions typically run only the resulting code and tools. Start with automation fundamentals, create a project folder in Visual Studio Code, install Claude Code, add a CLAUDE.md instruction file, and structure the project around workflows, an agent, and tools.

Transcript

Agentic workflows are not just a trend. They're the future of the AI industry. More and more companies are making the shift to agentic workflows. And this is just getting started because it's estimated that the AI agentic market is going from about $7 billion this year to around 93 billion in the next couple of years. So I can tell you right now th... Read More

Key Insights

• Agentic AI adoption is presented as a growing enterprise priority, with about 25% of enterprises deploying agentic pilots this year and an expected 50% doing so by 2027. The shift creates demand for workflow development, security planning, deployment knowledge, and ongoing system maintenance.
• Traditional automation is deterministic because developers explicitly map steps, connect nodes, and handle anticipated edge cases. It can fail when unexpected inputs or conditions appear, requiring someone to diagnose the issue, repair the workflow, and absorb the resulting maintenance cost.
• Self-healing behavior is available while an agent actively participates in a workflow run. In Claude Code, the agent can detect a failure, adjust its approach, modify its tools, and continue working, making this capability especially useful during development, iteration, and testing.
• Deployed workflows do not necessarily include the active development agent. When code and tools run automatically through a schedule or webhook, the agent's self-healing capability is absent, so the deployed system behaves more like a predictable traditional automation.
• The main advantage of agentic development is faster and more resilient construction. Instead of manually defining every connection, a builder can describe the desired outcome and let the agent implement details, respond to development problems, and account for edge cases that might otherwise be overlooked.
• Battle testing is essential before deploying an agent-built workflow. The tutorial compares this process to testing a train track with trains of different weights, lengths, and wheels, with the goal of confirming that the finished system handles varied conditions reliably.
• Automation fundamentals remain necessary when using Claude Code. Knowledge of APIs, webhooks, workflow logic, and manual automation helps builders evaluate generated code, identify poor decisions, describe requirements precisely, and determine whether the agent has produced a dependable system.
• A CLAUDE.md file acts as the project's persistent instruction layer. It tells Claude Code about folder structure, file locations, goals, and frameworks, while the WAT framework organizes development into workflows, an agent, and tools.

Questions & Answers

Q: What is an agentic workflow?

An agentic workflow is a multi-step automation built with assistance from an AI agent that can reason, make decisions, execute tasks, and adjust its approach during development. In the demonstrated setup, the agent works with project files and tools inside Claude Code. After deployment, the resulting code and tools may run without the active agent that originally helped build them.

Q: How do agentic workflows differ from traditional automation?

Traditional automation requires the builder to map each step, connect nodes or blocks, and anticipate edge cases manually. An agentic development process allows an AI agent to interpret an intended outcome, create the workflow, and respond to problems while building it. The deployed result can still be deterministic, particularly when only its code and tools run automatically.

Q: Can a deployed agentic workflow repair itself?

A workflow can adjust and recover while the Claude Code agent is actively involved in its execution. However, when the workflow is deployed to run through a schedule or webhook, the deployment typically contains the code and tools rather than the development agent. Its self-healing ability therefore goes away, and failures must be handled through the implemented automation logic.

Q: Why should agentic workflows be tested before deployment?

Agentic workflows should be battle tested because an agent's ability to solve problems during development does not guarantee that deployed code will handle every condition. Testing varied inputs and edge cases builds confidence in the deterministic automation. The tutorial illustrates this with multiple trains of different weights, lengths, and wheels testing the same track before deployment.

Q: Do I need traditional automation skills to use Claude Code?

Traditional automation skills remain valuable because they help you judge whether Claude Code made sound implementation decisions. Understanding APIs, webhooks, workflow logic, and common failure cases also improves how precisely you can describe the desired system. Without those fundamentals, a builder may accept generated work without recognizing structural problems or missing edge-case handling.

Q: How do I start building a workflow with Claude Code?

Start by downloading Visual Studio Code for your operating system and opening its extensions area. Find and install the Claude Code extension, then authenticate using a paid Claude subscription. Open a dedicated project folder, even if it is initially empty. Claude Code can then create and manage the files needed for the workflow inside that folder.

Q: What is the purpose of a CLAUDE.md file?

A CLAUDE.md file provides instructions for a specific Claude Code project and functions like a system prompt. Claude Code reads it together with user messages, allowing the file to define the project's goal, folder layout, file locations, and chosen frameworks. These persistent instructions give the agent clearer context whenever it plans, writes code, or modifies workflow components.

Q: What is the WAT framework for agentic workflows?

The WAT framework organizes an agentic project into workflows, an agent, and tools. Workflows define the processes being performed, the agent participates in planning and execution during development, and tools provide the capabilities used to complete tasks. In deployment, the workflow and tools can be released without deploying the active development agent itself.

Summary & Key Takeaways

• Agentic workflows address limitations in traditional automation by letting an AI agent reason through problems during development. The agent can revise its approach, update tools, and continue when errors arise. This assistance makes building faster, but the finished automation still requires careful testing before it can operate reliably without supervision.

• The tutorial sets up Claude Code inside Visual Studio Code. Users install the Claude Code extension, authenticate with a paid Claude subscription, open a dedicated project folder, and work with two primary elements: project files on the left and the Claude Code agent used for planning and execution on the right.

• A CLAUDE.md file supplies persistent project instructions that the agent reads alongside each user request. It can describe folder organization, file locations, project goals, and development frameworks. The demonstrated WAT framework separates the system into workflows, an agent, and tools, creating a clearer structure for building and testing automations.