Amazon Web Services has introduced a new IDE called Kiro, which uses advanced AI to help developers create software more easily. It’s built on a version of Visual Studio Code, so if you’re familiar with that, you’ll find it easy to get started. The idea is that you give Kiro a set of instructions or a project goal, and it helps generate requirements, designs, and task lists to build your app step by step.

How Kiro Works for Project Setup and Design

When you start a new project in Kiro, you’re prompted to give it a vibe or a formal spec. The vibe option is casual, good for quick ideas, while the spec is more detailed. For example, one user used the vibe to quickly create a Python script to check if virtual environments are valid. For more complex projects, like a static site generator, Kiro takes your prompts and produces a requirements document, a design outline, and a task list. These help guide the project from start to finish.

You can also create “steering” docs that set rules for how Kiro should interpret your instructions. These can include details about what the project should do, what tech stack to use, and how to organize files. You can generate or update these documents during the project, which then influence how Kiro refines its work. The requirements follow the familiar “user story” format, describing what the user wants and under what conditions. The design document details how the app’s components fit together, and the task list breaks down every step needed to build the project.

Guiding and Managing the Development Process

Kiro walks you through each step and provides real-time feedback on what it’s doing. It shows which files are being edited and what commands are being run. You can intervene at any time—changing code, adjusting commands, or redirecting the process. The system has an autopilot mode, but the reviewer chose to supervise each step to catch issues early. For instance, they had to manually specify that Python commands run with the “py” prefix on Windows, as Kiro didn’t automatically handle that.

Many of Kiro’s actions take a few minutes, so it’s designed to notify you when it needs attention. If you step away, it can resume its work later by re-reading the project documentation. However, this can sometimes lead to timeouts, especially if the AI API is slow or unresponsive. Interestingly, Kiro doesn’t seem to check for syntax errors before running code, which means some generated scripts contain mistakes that could be caught with simple linting—something it currently doesn’t do.

Automated Testing and Its Limitations

One of Kiro’s strong points is that it automatically writes unit tests as it develops code. It then tries to run these tests to verify correctness. When a test fails, Kiro attempts to fix it by rewriting the test or adjusting the code. For example, a test for a static site server initially failed because it didn’t account for stopping the server after testing. After some back-and-forth, Kiro suggested multiple fixes, including rewriting the tests entirely.

Despite these efforts, some tests still failed or produced strange results, like garbled text or missing templates. In one case, a sample blog post about best practices turned out to be nonsense. These issues highlight that, while Kiro can help automate parts of development, it still requires human oversight to catch inconsistencies or errors. Also, some generated code was more verbose than necessary, including features that weren’t explicitly asked for, like exporting data to JSON.

Overall, Kiro is an interesting step forward in AI-assisted software development. It tries to address common challenges like maintaining clear guidelines and iterative refinement. But it’s not perfect—limitations in AI context handling, occasional syntax errors, and incomplete tests show there’s still work to do. For now, it’s best used as an assistant rather than a fully autonomous builder, with developers keeping a close eye on its output.

Inspired by

• https://www.infoworld.com/article/4026617/first-look-guided-code-generation-with-kiro.html

Sources

• gmpg.org
• open-vsx.org
• siia.net
• podigee.com
• podigee.com