Hands-on with AWS Kiro's Agent Steering Feature

 

Demo Content & Objectives

Creating a Simple ToDo List

Ask AWS Kiro to Create a Simple ToDo List Application

Since we are only creating simple code this time, we will start with "Vibe Mode," which excels in intuitive operation.

 
Let's create a ToDo list with the following features:
- Add tasks
- Toggle completion/incompletion with checkboxes
- Delete tasks
- Save data in local storage (Retained even after closing the browser)

When I entered the request in the chat and executed it, it automatically created the script for the ToDo list app.
The chat also provides details about the implemented functions.

You can confirm that the created script files have been added under the folder in the top left of the screen.


Now, let's actually run the application.
When I asked "Run the application" in the chat, it automatically launched it in the browser.

I confirmed that the ToDo list app is implemented in the browser!
The features I requested were also implemented without any issues.

Setting up Agent Steering

Now, let's actually set up Steering. If you click the Kiro icon on the left side of the screen, several major functions of AWS Kiro will be displayed.
You should be able to see "AGENT STEERING" among them.

Click "Generate Steering Docs".

When clicked, a separate tab is created within the chat, and the workspace scan is executed.
Kiro reads the workspace and automatically creates the Product Overview (product.md), Project Structure (structure.md), and Tech Stack (tech.md).


Once the data reading is finished, you can verify that "product.md", "structure.md", and "tech.md" have been automatically created under the /steering directory.
The output was in English, which was hard for me to read, so I had it translated into Japanese.
I also want to check if it correctly interprets and implements instructions even in Japanese.

By the way, the contents look like this, and the files themselves can be freely edited.

Description of Each File

・【product.md】File
Defines the product overview of the workspace.
It defines the product's purpose, target users, key features, and business goals. This allows Kiro to understand the "why" behind technical decisions and propose solutions aligned with product goals.

・【structure.md】File
Defines the project structure of the workspace.
It outlines file organization, naming conventions, import patterns, and architectural decisions. This ensures that generated code fits seamlessly into the existing codebase.

・【tech.md】File
Defines the technical stack of the workspace.
It documents selected frameworks, libraries, development tools, and technical constraints. When Kiro proposes an implementation, the existing stack is prioritized over other options.


The Steering setup itself is now complete. However, this just read the script and defined rules based on its content. Let's add a new rule to define our unique project standards.

Adding Rules to Steering

Currently, the script created by Kiro does not contain any code explanations.
This makes it difficult to understand for third parties or when the code becomes complex.
Let's define a rule in Steering to include code explanations.

It is possible to edit the file directly, but since we are here, let's ask Kiro to add the definition.
I will add "Include code explanations in scripts" as a rule to the tech stack file.

It automatically corrected it to an appropriate format and appended it to the tech stack. Now, let's see if it actually gets applied.



Returning to the first tab, I will try executing a script modification to add a function to view deleted tasks in the ToDo list.
Without mentioning code explanations via comments, let's verify if the Steering rule applies.


It looks like the changes have been applied. It also outputs the files that were modified.


Let's check the modified script. Code explanations using comments have been appended to the script!
We confirmed that the Steering feature allows the AI to output answers that follow the project's conventions and definitions.

Now, it automatically creates them without me having to say "put in code explanations" every time.


By the way, if you want to check the changes in the modified file, you can compare it with the version before modification by clicking the left icon (the mark with arrows facing each other) from the modified file in the chat.


Since we changed the script, let's refresh the ToDo list and check. The deleted tasks are now visible!
The AI also took the liberty of adding other useful functions (restore, permanent delete, timestamps) that I didn't specifically mention.

Summary

How was it? In this post, we tried setting up Steering in AWS Kiro.

In traditional AI development, it was necessary to send instructions (commands) every time in the chat to control the response language and code style.
However, by defining project rules in advance using the Steering feature, the AI can now autonomously make optimal decisions without waiting for instructions.

While it is a very convenient feature, I got the impression that you won't receive much benefit if the definition settings are neglected.
I felt that Steering settings need to be defined carefully and in detail, firmly establishing the core parts of the project.

Next time, I plan to look into best practices and anti-patterns for defining settings in Steering!

↓At QES, we will actively disseminate information about Kiro, so please take a look!

 We aim to become Japan's #1 resource for Kiro! Our mission is to evangelize this technology by empowering end-users with top-tier education and technical support.