Kiro is an AI coding agent from AWS that plans before it codes. You describe a feature, and Kiro writes requirements, a design and a task list, together called a spec, before it builds anything.
The order matters. Many AI coding tools start writing code the moment you press Enter, while Kiro hands you a short document to check first, and that check is where the agent's wrong guesses get caught.
Why plan before the code
Give a coding agent a one-line prompt, such as 'add a checkout to my shop', and it has to fill in everything you didn't say. What happens when an item is out of stock? Which countries do you ship to? Does stock go down when the order is placed or when it's paid for? An agent that jumps straight to code picks an answer for each of those questions, and you only find out which one it picked by reading the code or by hitting the bug.
That's fine for a quick prototype, which is the spirit of vibe coding. It gets expensive on a real feature, where a wrong guess about one rule can mean undoing hundreds of lines.
Kiro's answer is spec-driven development: settle the decisions in plain-language documents first, then let the agent write the code from them. Correcting a sentence in a requirements file takes seconds. Correcting the same mistake once it's spread across ten files takes much longer.
How a spec works
A spec is a folder of three Markdown files that Kiro creates under .kiro/specs/ in your project, one folder per feature. You can commit them alongside the code, so the reasoning behind a feature stays in the repository.
Requirements
requirements.md says what the feature must do, as user stories with acceptance criteria. Kiro writes the criteria in EARS notation (the Easy Approach to Requirements Syntax), where each one reads 'WHEN something happens, THE SYSTEM SHALL do this'. For example: 'WHEN a customer adds more cans than are in stock, THE SYSTEM SHALL show how many are left.'
Wondering how this was made? How we make these