Scaling a One-Person Agency with Agentic AI Systems
Marketing Against the Graingo watch the original →
the gist
Barbara Jovanovic explains how she 5X'd her agency by moving from reactive AI prompting to a proactive, agentic system that integrates client context via MCPs and automated performance loops.
From Chat Box to Agentic Systems
Barbara Jovanovic transitioned her agency from using AI as a reactive chat interface to a proactive, multi-agent system. In 2025, she relied on manual prompting and copy-pasting; today, she utilizes Claude Code and agentic workflows to manage client operations. This shift allows her to maintain a one-person business while scaling output by 5X, handling everything from LinkedIn outbound to complex content production without additional hires.
Integrating Context via MCPs
The core of Jovanovic's system is the Model Context Protocol (MCP). By connecting tools like Granola (for meeting transcripts), Slack, and Google Drive directly into her AI environment, she ensures the model has deep, persistent knowledge of each client. This eliminates the need for manual context-loading. For clients without enterprise AI access, she uses a Slack-based workaround where meeting transcripts are automatically piped into dedicated channels, allowing her agents to ingest and index the data for future content generation.
The Founder Ghostwriter Workflow
Jovanovic argues that great founder-led content is not generated by generic prompts but by capturing authentic insights from the founder. She conducts monthly "insight calls" to extract controversial or unique industry perspectives. Her process involves:
- Data Ingestion: Automatically pulling transcripts from meetings.
- Dedicated Agents: Using specific agents for research, writing, and cold review.
- Cold Review Loop: A dedicated agent reviews the AI-generated content to catch "AI slop" and hallucinations before the human ever sees it.
- Performance Loop: A self-learning mechanism that refines future content based on past performance and feedback.
Why Process Beats Prompts
Jovanovic emphasizes that "prompts are dead" in favor of robust processes. Instead of relying on complex prompt engineering, she builds repeatable systems where the AI follows a structured workflow. This approach ensures consistency and quality, preventing the generic, "same-y" output common in AI-generated content. She notes that if a founder lacks genuine, interesting ideas, no amount of AI scaling will produce high-quality content; the AI's role is to amplify existing expertise, not invent it.