Stop building AI text wrappers. Build closed-loop systems instead. How we automated our agency.

If you are just plugging OpenAI into a text box and calling it a marketing tool, you are going to get crushed by churn.

A few months ago, my co-founders and I hit a massive wall. We were running a digital marketing agency out of Dhaka, manually grinding through B2B lead generation, editing videos, and writing SEO-optimized captions. We were throwing our own human hours at an operational nightmare.

Our clients—mostly high-ticket coaches and consultants—were exhausted. We realized that handing a client a viral piece of content is completely useless if their back-end pipeline is broken. Creating content, handling prospect objections, and posting across multiple platforms is draining.

As electrical engineering students, we couldn't help but look at this through the lens of circuit analysis. A marketing funnel shouldn't be a fragmented mess of ChatGPT, Google Sheets, and Hootsuite. It needs to be a closed-loop system. We realized a core truth: The winner with AI is determined by data quality and real workflows, not just AI access.

We threw out the idea of building another generic copywriter and focused entirely on a unified core loop: Scrape → Generate → Review → Publish.

Here is what we built under the hood to actually give founders their time back:

• Data-Driven Generation: Instead of generic prompts, our AI agent ('Ink') uses actual market trends and the specific objections founders hear on sales calls to draft platform-native content.

• The Human is the Editor-in-Chief: Pure automation ruins brand voice. We built a visual Kanban board where the AI pauses its workflow. You review the drafts, request specific edits, or hit approve with one tap.

• Auto-Publishing: Once approved, the engine automatically distributes the content, perfectly adjusting formatting and character limits for Instagram, LinkedIn, X, and TikTok.

The Tech Stack (For the builders here): We stripped out the overkill and focused on a rock-solid MVP.

• Orchestration: LangGraph (StateGraphs) + PydanticAI

• Backend: FastAPI handling REST and webhooks

• Database: Supabase (PostgreSQL + Realtime subscriptions for our Flutter dashboard)

• Publishing: Self-hosted Postiz instance managing all OAuth and scheduling queues

• Agents: Anthropic (Claude 3.5 Sonnet) handling the heavy cognitive routing

We launch later this week, and we are incredibly hungry to prove that a small, bootstrapped team can build world-class, globally competitive infrastructure.

For the solopreneurs and agency owners here: what does your current content and acquisition stack look like? Are you still manually bridging the gap between generation and publishing, or have you found a way to close the loop?

Would love to hear how you are handling this, or any harsh feedback on our approach!