How to Build and Sell AI Agentic Workflows for Business

TL;DR

Agentic workflows let AI agents complete multi-step tasks autonomously instead of being used as copy-paste tools. In a live demo, Claude Opus 4.5 found five Vancouver meal-prep companies, located their emails, and sent outreach messages in under 15 seconds, a task that would normally take 20 minutes. Learning this now captures an open arbitrage window.

Transcript

Hey, welcome to the definitive guide on agentic workflows for business. Now, agentic workflows have the potential to bring about what I think is one of the largest wealth transfers in human history. But very few people are currently talking about how to practically use them to improve their financial means. That's what this video is going to show y... Read More

Key Insights

• An agentic workflow is AI integrated directly into a business to complete multi-step tasks autonomously, not a 'glorified copy and paste tool' where users manually move answers between tabs, which the speaker calls a low-bandwidth, bottlenecked way of working.
• The AI 'overhang' is the gap between what large language models can actually do and what the general public believes they can do, meaning most people extract only a small percentage of AI's real value.
• Arbitrage is producing revenue or profit from a disparity in knowledge, and the window to profit from agentic workflows is wide open now but closing as more people learn how the technology works.
• Claude Opus 4.5 autonomously compiled five Vancouver meal-prep companies, searched their websites and contact pages for emails, and sent outreach messages, turning a 20-minute task into one taking under 15 seconds.
• MCP, which the speaker calls a model contact protocol server, is what let the agent actually send the outreach emails, and it was set up in advance to give the agent that capability.
• The DO framework stands for directive, orchestration, and execution, and is presented alongside Claude skills, MCP, and other frameworks as tools to structure agentic workflows for different situations.
• Self-annealing workflows heal themselves when they error out, which is one of several reliability concerns the course addresses along with testing, validation, and troubleshooting when things break.
• Agentic workflows should be viewed as augmenting parts of a role rather than automating 100 percent of a role, correcting the common but inaccurate belief that whole teams are simply being replaced.

Questions & Answers

Q: What is an agentic workflow?

An agentic workflow is AI integrated directly into a business to autonomously complete multi-step tasks, acting as an extension of yourself. The speaker contrasts this with how most people use AI as a 'glorified copy and paste tool,' asking questions and manually moving answers from one tab to another, which he calls a low-bandwidth, bottlenecked way of working. Agentic workflows instead have the model search, gather data, and take actions on its own without requiring input at each step.

Q: What is the AI overhang?

The AI overhang is the gap between the reality of what AI is capable of and what the general public believes or knows how to use. The speaker illustrates it as two lines: a lower one representing public perception and ability, and a higher one representing reality. Because of this gap, most people are only squeezing out a very small percentage of the actual value of large language models and agentic workflows, treating them as external third-party tools rather than integrated systems.

Q: How did the Claude Opus 4.5 meal-prep demo work?

The speaker asked Claude Opus 4.5 to compile five local meal-preparation companies delivering to downtown Vancouver, find their email addresses, and email each with his specifications of roughly 3500 calories and 200 grams of protein per day, entirely autonomously. It searched, found companies like Fresh Prep, Two Guys with Knives, and WellFed, broadened its search when emails were scarce, accumulated addresses in a temporary database, and sent five emails via an MCP server in under 15 seconds.

Q: What does MCP mean in this video?

The speaker refers to MCP as a 'model contact protocol server' that he had set up in advance. In the demo, MCP is what allowed the agent to actually send the outreach emails to the meal-prep companies, rather than just draft them. He describes it as one of several frameworks, alongside Claude skills and the DO framework, that the course explains in terms of what each does, when to use which, and how they fit together.

Q: What is the DO framework?

The DO framework stands for directive, orchestration, and execution. It is introduced as part of the building portion of the course, listed among the tools the speaker teaches for structuring agentic workflows, including Claude skills, MCP, and other frameworks. The course promises to explain what each of these does, when to use which one, and how they all fit together, so that unstructured agent requests can be wrapped in a proper framework for more reliable results.

Q: Why is now considered an arbitrage opportunity for agentic workflows?

Arbitrage is the ability to produce a beneficial outcome, revenue, or profit based on a disparity in knowledge. The speaker argues that because he and few others understand agentic workflows while most of the market does not, there is a gap the market will pay to close. That window is closing as people learn how the technology works, but right now it is wide open, and he claims you can make a large amount of money exploiting this knowledge gap.

Q: What does it mean for an agentic workflow to be self-annealing?

Self-annealing means the workflow can heal itself when it errors out, recovering without manual intervention. The speaker lists making workflows self-annealing as one of the course topics, alongside testing and validating workflows, choosing the best system prompts, and troubleshooting when things break. This reliability focus reflects that agentic workflows encounter common problems, and the course is designed to teach both how to build them and how to fix them when they fail.

Q: Who is the instructor and what is his background?

The instructor is Nick, who runs LeftClick, an AI growth agency that has done work for large brands like Mr. Beast, Anthropic, and OpenAI. He says he built two AI-based service agencies to $160,000 a month in combined revenue and consulted for a couple of billion-dollar businesses. His family immigrated from the People's Republic of Bulgaria in the 90s, he has no programming degree, learned everything from YouTube, and started his first business at 19.

Summary & Key Takeaways

• The course teaches agentic workflows with a business focus, promising to cover what they are, how they function via loops, common problems and fixes, and hands-on building including workspace setup, first flows, the DO framework, Claude skills, MCP, testing, system prompts, and cloud deployment.

• The speaker, Nick of LeftClick, built two AI-based service agencies to a combined $160,000 per month and consulted for billion-dollar businesses. He frames agentic workflows as one of the largest wealth transfers in history and an open arbitrage opportunity right now.

• A live demo showed Claude Opus 4.5 autonomously finding five Vancouver meal-prep companies, gathering emails from contact pages, and sending outreach emails via an MCP server in under 15 seconds, with output the speaker says matches his own tone without detectable AI-isms.