This AI agent is my new marketing department (AgentKit + FeedHive MCP)

Here's a detailed summary of the YouTube video transcript:

Key Concepts

• AI Agents: Specialized AI programs designed to perform specific tasks within a larger workflow.
• OpenAI Agent Builder & Agent Kit: Tools for creating and orchestrating AI agents.
• Web Search Tool: An AI tool that allows agents to access and retrieve information from the internet.
• Reasoning Effort: A setting that controls the complexity of thought an AI agent can employ.
• User Approval Step: A checkpoint in a workflow where a human reviews and approves the AI's output before proceeding.
• Brand Brief: A document that outlines a brand's voice, tone, values, and target audience, used to guide AI content generation.
• Fine-tuning: The process of adapting a pre-trained AI model to a specific task or style.
• MCP Server (Meta-Communication Protocol): A protocol that enables different AI systems and tools to communicate and exchange data.
• FeedHive: A social media management tool that integrates AI for content creation and scheduling, including a custom MCP server.
• Lovable: An AI tool that builds software interfaces based on user prompts and documentation.
• Founder Stack: A collection of five SaaS tools for bootstrap founders, including FeedHive, offered with lifetime access and AI credits.

Building an AI-Powered Content Creation System

This video outlines a step-by-step process for building an automated system that leverages AI agents to identify news, research it, and draft social media posts in a unique brand voice, ultimately aiming to establish users as "top voices" in their industries.

1. The Vision: From News Hunting to Publishing

The core problem addressed is the exhaustion of constantly chasing big stories and the difficulty of building authority through timely content. The proposed solution is a two-AI agent system:

• AI Agent 1 (News Scout): Scours the news for relevant topics.
• AI Agent 2 (Personal Ghostwriter): Drafts posts in the user's unique voice. This system aims to shift the user's role from active content creation to simply approving and publishing, ensuring consistent presence at the forefront of industry conversations.

2. Setting Up the Core AI Agents with OpenAI's Agent Builder

The foundation of the system is built using OpenAI's Agent Builder and Agent Kit. The workflow is broken down into distinct AI agents, each with a specific responsibility and access to necessary tools.

2.1. The "News Scout" Agent

• Instructions: To search for news related to a user-provided topic and present a list of headlines.
• Tool: Web Search.
• Configuration: Broad and open search, medium context size.

2.2. The "News Reporter" Agent

• Instructions: To select a suitable headline from the Scout's output, research it further, and create a news report. It should also find a relevant image.
• Input: Output from the "News Scout" agent.
• Tool: Web Search.
• Configuration: Wide open search, medium search context size, medium reasoning effort (due to the need for more complex thinking).
• Context Management: The "include chat history" option is disabled to maintain a clean context.

2.3. User Approval Step

• Purpose: To provide a checkpoint for human review of the news report and image before proceeding. This is a temporary step for testing and can be removed later.

2.4. Initial Testing and Output

Upon testing, the system successfully:

1. Searches for news.
2. Presents a news report based on a chosen headline.
3. Finds a relevant image.
4. Asks for user approval to continue.

3. Crafting a Unique Brand Voice: The Challenge of AI Ghostwriting

The video emphasizes that simply dumping raw news reports into a general AI like ChatGPT will result in "sloppy" and easily detectable AI-generated content that can harm a brand. True AI ghostwriting requires the AI to deeply understand the user's:

• Brand
• Specific writing style
• Tone of voice
• Brand values

This level of personalization is complex and typically involves fine-tuning AI models, creating comprehensive brand briefs, and implementing quality assessment loops.

4. Streamlining Brand Voice Integration with FeedHive

The presenter introduces FeedHive as a solution that streamlines the process of integrating a unique brand voice into AI-generated content.

4.1. FeedHive Setup

• Social Account Connection: Users connect their social media accounts.
• Brand Profile:
  • Writing Style Selection: Users choose a pre-defined writing style that best matches their desired output (e.g., "vertical writing style" for LinkedIn).
  • Brand Brief Generation: FeedHive analyzes connected social accounts and personal websites to automatically generate a comprehensive brand brief. Users can edit this brief.
• AI Post Creation Test: A quick test on the "compose" page demonstrates that the AI can generate posts that sound authentic to the user's brand.

4.2. Automation Trigger in FeedHive

• Trigger Creation: An "AI News Trigger" is created to publish for LinkedIn and Twitter.
• Action: The trigger is set to "create draft" for now, allowing for review before publishing.
• AI Model Selection: GPT-5 is chosen for writing the post.
• Output: FeedHive is now configured to take a prompt and combine it with the user's brand brief and writing style to produce a ready-to-publish social media post.

5. Integrating FeedHive with OpenAI's Agent Builder

The next step is to connect the FeedHive automation to the OpenAI agent workflow.

5.1. MCP Server Integration

• New Agent Node: An additional agent node is added after the approval step.
• Tool: MCP Server.
• Configuration: The FeedHive MCP server URL (mcp.feedhive.com) is added.
• API Key: An API key is generated from FeedHive's settings and pasted into the OpenAI agent builder to establish a connection.
• Trigger Recognition: The "AI News Trigger" created in FeedHive appears as a tool within the OpenAI agent builder.

5.2. The "Prompt Agent"

• Purpose: This agent's job is to create a prompt for FeedHive and pass along the news report and image.
• Instructions: It's instructed to use the FHMCP server tool.

5.3. Final Workflow Structure

The complete workflow consists of:

1. News Scout Agent
2. News Reporter Agent
3. User Approval Step
4. Prompt Agent (using FeedHive MCP Server)

6. End-to-End Testing and Final Output

• Prompting: The system is tested with a prompt (e.g., "OpenAI making a deal with AMD").
• Execution: The agents execute their tasks sequentially.
• Prompt Agent Output: The prompt agent breaks down the prompt's components.
• FeedHive Draft: The final output is a draft post in FeedHive, complete with a well-written news report and an image (e.g., of Sam Altman).
• Scheduling: The user can then schedule this draft for publication.

7. Alternative Approaches and Tool Synergy

• NAD (Not Another Decision): The presenter acknowledges that NAD could also be used for this workflow. However, for a chat-based starting point, OpenAI's Agent Builder is considered more promising.
• Agent Builder and NAD Synergy: For more complex workflows, Agent Builder can delegate tasks to NAD, which can then delegate to tools like FeedHive. The tools are seen as complementary rather than competing.

8. Creating a Standalone Chat Interface with Lovable

To enhance user experience, the workflow can be published and accessed via a standalone chat interface.

• Publishing the Workflow: The workflow is published, and its ID is obtained.
• Lovable Integration: The workflow ID and documentation are provided to Lovable, an AI tool that builds software interfaces.
• Result: Lovable generates a user-friendly chat interface accessible via a website, allowing for easier interaction with the AI system.

9. Founder Stack Offer

The video concludes by promoting the "Founder Stack" offer, which provides lifetime access to FeedHive and four other AI tools. This includes:

• 50,000 AI credits per month.
• The option to use a personal OpenAI API key for unlimited credits.
• A single, one-time purchase.

An alternative video is recommended for users who prefer not to use FeedHive, detailing how to replicate the workflow with other social media tools.