TLDR;
This video explains why agentic workflows are the future of AI and demonstrates how to build one using Cloud Code. It covers the shift towards agentic systems, the limitations of traditional automation, and the advantages of using agentic workflows. The video also provides a live build of a newsletter automation workflow, showcasing how to use Cloud Code, and discusses how to sell these skills for premium prices.
- Agentic workflows are predicted to grow significantly in the AI market.
- Traditional automation has limitations that agentic workflows can overcome.
- Cloud Code simplifies the creation of agentic workflows, even for non-developers.
- Understanding business needs and providing value-based solutions is key to selling AI automation skills.
Intro - Why now? [0:00]
Agentic workflows are identified as the future of the AI industry, with the market expected to grow from $7 billion to $93 billion in the next few years. Companies are increasingly adopting agentic workflows because they realize they can move faster than with traditional automation. Knowing how to build these workflows is becoming a valuable skill.
Big money moving in [0:38]
The agentic AI market is currently around $8 billion and is projected to reach $40 to $50 billion by 2030. Approximately 25% of enterprises are deploying agentic pilots this year, and this number is expected to rise to 50% by 2027. This shift is driving massive budget allocations and creating new opportunities for those skilled in building these systems.
Why automations break [1:15]
Companies are reaching the limits of traditional automation and are turning to agentic workflows for faster progress. Traditional workflows, built with tools like Eniden or Zapier, require manual mapping of every step and break when encountering unexpected issues, leading to maintenance costs and time.
“Self-fix” - does it work? [2:10]
Agentic workflows have self-healing capabilities, especially during the building and iteration phases within tools like Cloud Code. However, once deployed to run independently, the self-healing ability diminishes, and the system behaves more like traditional automation. This predictability is beneficial, as automations are more reliable and deterministic. The real advantage lies in the building process, where the agent handles edge cases and ensures a more robust final product.
Why basics still matter [3:50]
Understanding the fundamentals of traditional automation is crucial for working with agentic workflows. Those who skip these basics may struggle to assess the quality of what the agent builds and may lack the ability to spot errors or communicate effectively with the agent. A solid understanding of automations allows for clearer communication and better outcomes when using tools like Cloud Code.
Claude Code setup [4:28]
To build an agentic workflow using Cloud Code, you need Visual Studio Code and a paid Claude subscription. After installing the Cloud Code extension and authenticating, you open a project folder in Visual Studio Code. Cloud Code operates with agents and files, where the agent interacts with the user to plan and execute code. A claude.md file, acting as a system prompt, provides instructions for the project, including folder layouts, goals, and frameworks like WAT (workflows, agent, and tools).
Control the Agent [7:18]
The WAT framework includes workflows (natural language processes) and tools (ingredients). Workflows guide the agent on how to use the tools, and both improve over time as the agent learns. Setting the agent to plan mode allows for clear communication and refinement of the workflow before execution. Providing specific instructions and brand assets helps the agent create tailored outputs.
Live build starts [10:09]
The live build demonstrates creating a newsletter automation workflow using Cloud Code. The agent is instructed to research, structure the newsletter in HTML, and create infographics. The agent asks clarifying questions and generates a plan, which includes using perplexity for research, Claude for content, nano banana for infographics, and Gmail for delivery. Brand assets are added to ensure the newsletter aligns with brand guidelines. The agent then generates config files, tools, and a workflow, and the user adds API keys to the .env file. The agent encounters and fixes a unicode encoding issue and updates the tool with the correct endpoint. The final newsletter includes branded elements, AI-generated images, and key takeaways.
How to sell it [20:56]
To sell AI automation skills, focus on identifying and solving business constraints rather than just building flashy demos. Act as a doctor who diagnoses problems and prescribes solutions, not just a pharmacist who fills prescriptions. Lead with the value you can provide, such as saving time or reducing errors, and price your services based on the value delivered, not just the time spent. Building a $3,000 project can turn into a $50,000 a year relationship by tracking metrics and proactively showing the value of the system.
