The "Product Manager" in the Machine
As a Product Manager, my day-to-day usually involves roadmap alignment, user interviews, and prioritization. However, in my "off-hours," I’ve become obsessed with the potential of Agentic Development. I’ve spent months building small functional solutions for my own workflow and helping colleagues automate their business processes.
When the DataArt Agentic Development Hackathon was announced, I saw it as the ultimate test. Could I take my hobbyist experimentation and apply it to a high-stakes, time-sensitive competitive environment?
The answer lay not in how well I could code, but in how well I could orchestrate.
The Vision: A Modern "El Chat"
The prompt for the hackathon was clear: Create a functional chat server. Immediately, my mind went back to the early days of the internet—specifically to a site called elchat.com. For many of us in Latin America, it was a "rusty" but beloved digital plaza where you could connect with anyone, anywhere.
My goal wasn't just to build a server; it was to build a modern, agent-driven version of that nostalgic experience. But as a PM, I didn't reach for a compiler first. I reached for my User Stories.
Phase 1: Hiring my "Senior Business Analyst" (Gemini)
To succeed in a 48-hour sprint—especially while living in Central Time and balancing a different schedule than my international peers—I needed a solid foundation.
I started by crafting a highly specific Profile Prompt for Gemini. I didn't just ask it for ideas; I commanded it to act as a Senior Business Analyst. We entered an intensive back-and-forth dialogue:
- Refining the core features.
- Identifying potential edge cases.
- Breaking down the roadmap into a lean, 2-day "hackathon" sprint.
By the end of the session, I had a professional-grade Product Backlog. I knew exactly what the agents needed to build before a single line of code was generated.
Phase 2: Spec Driven Development (SDD)
Once I opened VSCode, I committed to a Spec Driven Development approach.
For a Product-oriented person, SDD is a game changer. It shifts the focus from the "boring" mechanics of writing syntax to the high-level strategy of defining requirements. It felt like I was managing a development team that never slept. I provided the specs; the AI provided the delivery. This allowed me to stay focused on prioritization—ensuring that the most critical features were built first.
Phase 3: Building the "Agentic Squad"
One of the most exciting parts of this hackathon was the requirement for Dockerization. Every solution had to be packaged in a Docker container with zero external dependencies. This meant the architecture had to be tight, portable, and self-contained.
To manage this, I didn't just use one AI. I built a squad:
- The Developer Agent: Handled the core logic and feature implementation.
- The Backend Tester: A specialized agent focused on API reliability and server-side stability.
- The Frontend Tester: Ensuring the UI was responsive and user-friendly.
The "Frustrating" Reality of Agentic Work
I’ll be honest: there were moments of deep frustration. Even though I’m not a professional developer, I had a burning urge to get in there and manually "touch the code."
However, the "Agentic" philosophy requires a different kind of discipline. I realized that my value wasn't in fixing a typo in a CSS file; my value was in refining the prompt, adjusting the spec, and ensuring the testers were catching the right bugs.
The Takeaway: The Human in the Loop
At the end of the hackathon, I walked away with a functional, containerized chat server and a profound realization.
Even when AI takes over 90% of the development work, the 10% that remains—the decision-making, the strategy, and the vision—is more critical than ever. An AI can write a function, but it cannot decide why that function matters to the user. All the decisions behind the code must come from a human with the experience to know what "good" looks like.
Final Credits
A huge thank you to DataArt for the opportunity to push the boundaries of what "development" means in 2024. To my fellow participants: thumbs up for the incredible work and the shared late nights.
🔗 Explore the Project
Curious about how it turned out? You can find the full repository, including the Docker setup and the agent-generated logic, right here:
Top comments (0)