Inspiration
We were inspired by the gap between fast AI code generation and robust, maintainable software. Many AI coding tools produce quick snippets but skip clear specifications, validation, and traceability. Kiro’s spec-first agentic approach caught our eye as a promising way to bring engineering discipline into AI-assisted development. We wanted to test whether this structured workflow can make AI-built projects easier to reason about, verify and iterate on in a hackathon setting.
How we built it
We built the whole project using Kiro’s agentic workflow. We started by writing clear specs to define requirements and acceptance criteria before writing any implementation code. Kiro broke down the tasks, generated implementation plans, produced code, and ran validation checks against our original spec. We reviewed each iteration, adjusted requirements, and guided the agent to refine output. We kept all changes traceable and aligned with the initial specification throughout the hackathon.
What we learned
We learned that the spec-first model fundamentally changes how you collaborate with an AI agent. Writing precise specifications becomes the core skill, rather than just prompting for code. We also saw how helpful built-in validation is for catching logic gaps early. This workflow encourages us to think about edge cases, test coverage and maintainability from the very beginning, not as an afterthought.
Challenges we faced
Our biggest challenge was learning how to write effective specs. Vague requirements led the agent to produce ambiguous or overcomplicated implementations. We also had to balance hackathon speed with the rigor of the spec process — it took extra time to formalize plans upfront. Iterating on specs and refining acceptance criteria was slower at first, but it reduced costly rewrites later in the project.
Built With
- agent
- agentic
- ai
- machine-learning
