AI Coding Hackathon Winners Revealed LIVE ($17K in Prizes)

Key Concepts

• AI-Powered Hackathon Success: The Dynamus & Hero AI hackathon generated 160 high-quality submissions, showcasing the potential of AI-assisted coding.
• Kuro & Claude as Core Tools: Kuro and Claude Code were central to many projects, with judging criteria heavily weighting their effective utilization (sub-agents, prompts, hooks, agent teams).
• Importance of Holistic Evaluation: Judging prioritized not just application “coolness” but also code quality, documentation, innovation, and real-world utility.
• Agentic Coding Workflow: A structured workflow involving planning, implementation, and rigorous testing is crucial for successful agent-based development.
• Planning as a Critical Phase: Thorough planning, including interactive questioning with the AI, minimizes assumptions and improves code quality.
• End-to-End Testing Imperative: Comprehensive end-to-end testing, ideally automated, is essential for ensuring application reliability.
• Community & Learning Resources: The Dynamus community and associated courses provide valuable resources for learning and mastering agentic coding techniques.

Hackathon Overview & Winner Announcement (Part 1 & Initial Part 2)

The Dynamus and Hero AI coding hackathon attracted 160 submissions, making the judging process challenging. The $17,000 prize pool was distributed as $5,000 (1st), $3,000 (2nd), $2,000 (3rd), and $1,000 each for 4th-10th place. Judging criteria included application quality, Kuro usage (sub-agents, prompts, hooks), documentation (devlogs), innovation, and presentation. The host emphasized the high quality of all submissions, regardless of placement. A webpage will showcase all projects, with live demos of the top three. Notable projects mentioned included Open Cloth Repos (100,000 stars), Zsign (self-hosted DocuSign alternative), and Mega (AI-powered print-on-demand automation).

The initial winners announced included Ahmed (10th place, Channel Chat utilizing RAG architecture to ingest and search channel content), and the second-place winner, whose project was defended by the judges despite its simplicity, citing effective Kira usage and practical application. The judging process was defended against criticism from the live chat.

First Place Winner & Project Deep Dive (Later Part 2)

Jamal’s “Live Stock AI Manager” was announced as the first-place winner, receiving $5,000. This application is a feature-rich, offline-first livestock management system deployed on Cloudflare Workers, supporting multiple species and languages. Its success was attributed to effective Kira AI usage, specifically context engineering within the agent system, utilizing multiple agents with tailored file access for internationalization and livestock specialization. Uniquely, the project integrated Kira AI as a “second brain” directly within the codebase, allowing developers to query the code via a CLI agent.

Live Coding Demonstration: ChargeB Integration (Part 3 & Initial Part 4)

The segment transitioned to a live coding demonstration using Claude Code’s new “agent teams” feature. The goal was to build a ChargeB payment integration into an existing chat application, utilizing Superbase for authentication. The demonstration emphasized a structured planning phase, beginning with a “prime” command to familiarize the agent with the codebase, followed by prompting it to ask clarifying questions. The importance of reducing assumptions made by the coding agent was repeatedly stressed.

The initial attempt to leverage agent teams was hampered by resource constraints (RAM), leading to a crash. However, the speaker continued, highlighting the value of a detailed plan and advocating for a “contract-first” approach, defining clear communication protocols between agents (front-end, back-end, database). The speaker also demonstrated the use of the Vercel Agent Browser CLI for end-to-end testing.

Testing, Refinement & Tool Discussion (Later Part 4)

The live coding session continued with a focus on testing and refinement. While Claude Code is considered the most powerful coding agent, it was noted to be “insanely bloated and overengineered,” leading to inefficiencies and memory leaks. A recurring issue with Opus 4.6 was identified: it consistently skipped the end-to-end testing phase, requiring manual intervention. Opus 4.5 was noted as performing this task more reliably. Prompt engineering was highlighted as crucial, with newer models requiring adjustments to prompts.

The speaker cautioned against using OpenClaw due to significant security vulnerabilities, advocating for building custom, secure second brains using tools like Cloud Code. The Dynamus community and agentic coding course were heavily promoted, offering a 20% discount. The speaker reiterated that Cloud Code generally produces better code quality than Kuro, recommending Kuro for proof-of-concepts but Cloud Code for overall output.

Conclusion

The Dynamus & Hero AI hackathon and subsequent live coding demonstration showcased the transformative potential of AI-assisted coding. While tools like Kuro and Claude Code offer significant advantages, success hinges on a structured workflow emphasizing thorough planning, rigorous testing, and a deep understanding of the underlying technologies. The Dynamus community and associated learning resources provide a valuable ecosystem for developers seeking to master these techniques and unlock the full potential of agentic coding. Despite current limitations, the future of AI-powered development appears promising, with ongoing advancements in agent teams and other innovative tools.