I’m Automating a Business LIVE to Prove Anyone Can Do It (Part 2)

Key Concepts:

• MVP (Minimum Viable Product): A version of a product with just enough features to satisfy early customers and provide feedback for future product development.
• Lead Qualification: The process of determining whether a potential customer is a good fit for a business's products or services.
• API (Application Programming Interface): A set of rules and specifications that software programs can follow to communicate with each other.
• MCP (Managed Code Platform): A platform for building and deploying custom tools and agents.
• Observability: The ability to understand the internal state of a system from its external outputs.
• AI Workflow: A sequence of automated tasks powered by artificial intelligence.
• Agents: Autonomous entities that can perceive their environment, make decisions, and take actions to achieve specific goals.

1. Explore Stage: Reviewing Opportunities and Researching Implementation

• The video focuses on automating two key processes in Steven Clark's wholesaling business: the Google form fill-out process and the quotation process/offer confirmation.
• The initial step involves exploring potential APIs for property valuation.
• ChatGPT is used to identify suitable APIs, specifically for determining home values.
• The Zillow API is noted as unavailable.
• Rentcast is selected as the preferred API due to its reasonable pricing and comprehensive documentation.
• The "value estimate endpoint" in Rentcast is identified as crucial for the agent to determine offer suitability.

2. Building the Solution: Utilizing Agency and Custom MCP Server Template

• The solution is built using the presenter's platform, Agency, with a custom MCP server template.
• The template facilitates rapid building and deployment of custom MCP servers.
• The template includes a GitHub repository that allows creating MCP servers without extensive coding.
• The Rentcast API documentation is copied and used as input for Cursor, an AI-powered code editor.
• Cursor is prompted to create a tool that utilizes the Rentcast API.
• The Rentcast API key is added to the environment file.
• The tool is tested and successfully retrieves a list of comparable properties.
• A second tool is created to analyze transcripts from sales calls using Whisper for transcription.
• Cursor is prompted to create a tool that transcribes audio from a given URL.
• A real audio file from the client's CRM is used to test the transcription tool.
• The transcription tool initially encounters an error, which Cursor automatically fixes.

3. Deployment: Using Railway and Integrating with Agency

• Railway, a modern infrastructure platform, is used to deploy the tools from the GitHub repository.
• Deployment is initiated by pushing changes to GitHub and then selecting the repository in Railway.
• A domain is generated in Railway for accessing the deployed tools.
• The deployed tools are integrated into Agency, ensuring the client's account is used for easier modifications.
• The server URL from Railway is added to Agency, allowing the platform to recognize the deployed tools.
• The property valuation tool and audio transcription tool are created separately within Agency.
• Authentication can be added by including an app token in the environment file.

4. Agent Creation: Real Estate Evaluator and Transcription Agent

• Two agents are created: a "Real Estate Evaluator Agent" and a "Transcription Agent."
• The Real Estate Evaluator Agent uses the property valuation tool.
• A prompt from a previous attempt by the client to create a custom GPT is used as a starting point.
• The agent is asked what is confusing in its own instructions to identify areas for improvement.
• The prompt is refined based on the agent's feedback.
• Example data is used to test the Real Estate Evaluator Agent.
• The agent successfully uses the tool to provide property valuation, price recommendations, and an overall rating.
• The Transcription Agent is designed to transcribe calls and generate meeting summaries.
• OpenAI FM is used to generate a dummy script for testing the Transcription Agent.
• The agent successfully generates details and a brief summary of the call from the dummy script.

5. Integration: Zapier and Go High Level (REI Reply)

• The agents are integrated into the client's existing systems: Monday CRM (for form data) and REI Reply (Go High Level) for lead qualification.
• Zapier is used to integrate the Real Estate Evaluator Agent with Monday CRM due to limitations in Monday's workflow features.
• The Agency app is installed on Zapier.
• The "prompt" event is selected in Zapier, and the Agency account is connected.
• Fields from the Monday CRM event are mapped to the agent's input parameters.
• The output from the agent is saved to a Google Chat space for communication.
• The Go High Level CRM is integrated using a custom web hook.
• The trigger is set to "call status completed."
• The action is a custom web hook with the agent's URL and parameters.
• The call recording URL is passed as a custom value.
• A note is added to the lead with the response from the agent.
• Go High Level's built-in call transcription feature is used as an alternative to accessing private recording URLs.
• The call transcript is passed to the agent for analysis.

6. Evaluation: Observability and Linkfuse

• Evaluation is crucial to assess the performance of the agents.
• The platform has observability integrated with Linkfuse.
• The video references a podcast episode with Adam Silverman on observability.
• The next part of the series will focus on evaluating agent performance and making adjustments.
• The goal is to remove human intervention once the agents are performing as expected.
• Future videos will explore more complex agents, such as voice agents.

7. Notable Quotes:

• "Today, agents understand themselves better than we do." - This highlights the importance of leveraging AI's self-awareness for prompt engineering.
• "...you don't have to focus on only agents. So if your client currently requires an AI workflow, if this is what they need the most right now, then build this for them and then transition to more sophisticated and general agents later." - This emphasizes a pragmatic approach, prioritizing immediate client needs over complex solutions.

8. Technical Terms and Concepts:

• Temperature: A parameter that controls the randomness of the agent's output. Lower temperatures result in more deterministic and predictable responses.
• Prompt Engineering: The process of designing effective prompts to guide the behavior of AI models.
• Bearer Token: A security token used to authenticate API requests.
• Web Hook: An automated HTTP request triggered by an event in one system and sent to another system.

9. Logical Connections:

• The video builds upon the previous one by taking the identified opportunities and moving into the MVP development phase.
• The explore stage logically leads to the building stage, which then leads to deployment and integration.
• The integration phase sets the stage for evaluation and further refinement.

10. Synthesis/Conclusion:

The video details the process of building an MVP for automating key processes in a wholesaling business using AI agents. It covers API exploration, tool creation using Agency and Cursor, deployment with Railway, integration with existing CRM systems (Monday and Go High Level), and the importance of observability for evaluation. The emphasis is on a practical, iterative approach, starting with simple AI workflows and gradually transitioning to more sophisticated agents. The video highlights the importance of understanding client needs and leveraging existing tools and platforms for efficient development and deployment.