🤖 AI Up-Skilling Journey | Experiment 2: Social Media Content Generator

🎯 Goal:

Learn Open AI's Agent Builder by designing a workflow that ingests unstructured notes about an AI experiment and automatically produces:

• a structured website summary,

• a LinkedIn post,

• an Instagram caption,

• and a visual prompt for an AI-generated image.

🛠️ Tools

• OpenAI Agent Builder for multi-agent orchestration

• ChatGPT (4o-mini) for reasoning and content generation

• Rube MCP as a Notion read/write bridge

• Notion as the source of truth and content store

• Gemini for AI image generation

🗺️ Architecture

I started with one monolithic agent and quickly ran into complexity and inconsistency. The final design uses several focused agents instead, each with a clear job: parsing and improving notes, generating channel-specific content, creating image prompts, and integrating with Notion.

🧠 Key Learnings

• Multi-agent designs work better than a single agent when the workflow has distinct stages.

• Rube MCP is an effective workaround for Notions OAuth-only MCP in environments that prefer API keys.

• Newer models can hit transaction limits quickly; 4o-mini gave a better balance of cost, stability, and performance.

• Human-in-the-loop oversight is essential to counter hallucinations, wrong tool suggestions, and over-engineered plans.

• Systems thinking especially being deliberate about which agents can call which tools and when keeps the workflow maintainable.

🏁 Outcome

The workflow can now take my rough notes, clean and structure them, generate tailored content for web and social channels, and produce an aligned image prompt while staying within transaction limits.

#AgentBuilder #Notion #Rube #AgenticSytem #OpenAI #ChatGPT #AI