How far Kiro got me in 2.5 Hours |Kiro is an agentic AI IDE

I attended an Adelaide AWS User group Kiro Night on Feb 12 2026. Kiro is an agentic, AI-powered Integrated Development Environment (IDE) from AWS based on VS Code. We did lots of awesome stuff after being shown around Kiro by Anton Schnetler and Arran Peterson from AWS, like matrix themed glyph animation and even our own space invaders application. This stuck with me.

So I ran a simple experiment.

How far can I get building a real, end-to-end data product using Kiro — without writing code myself?

Q – Data NERD

Constraints

no manual coding (even in my native coding languages)
focus on time to delivery, not polish
work within my real-world constraints – shared hosting, MySQL, PHP

The Outcome

Q’s LEGO Collection

Distribution of sets by theme (percentage of total)

Approximately 2.5 hours later, I completed the experiment phase (details below). I stopped myself from over-polishing, but there was a bit of post experiment tweaking between the completed experiment (23rd March) and today.

EXPERIMENT PHASE

credits 13 .25

prompts 20

API ingestion (Rebrickable)
Authentication working
A backend API layer (PHP)
An Interactive Visualisation embedded in my site

TWEAKING PHASE

credits 7 .21

prompts 8

Deliberately deferred visual refinements

What worked Well

no syntax issues or broken code
Kiro handled structure across Python, SQL, PHP and javascript.
Once the direction was clear it executed reliably

Iteration wasn’t about fixing broken code, it was about changing requirements.

Some Friction Points

Configuration settings, while iterating prior to sanitising hard coded config was repeatedly overwritten. I did catch this early and started saving the hardcoded blocks separately for re-pasting.
Frequently (I would say up to quarter of the prompts), I would submit the prompt, Kiro would respond with “Understood.” Use some credits and not actually edit any code or files.
Sometimes… like with people, we misunderstood each other. At one point I asked to add a naming convention to the SQL steering file and apply it to objects in the database, Kiro applied it to my .sql files.
When I followed the getting started documentation, one of the first thing you request is for Kiro to generate steering files. I was starting in an empty folder and feel like It would have been better to generate at least a readme prior to generating the steering files.

What I still did without Kiro

Running DDL (data definition language) scripts in the database.
Uploading files to the host when editing was completed.
Navigating hosting constraints (most popular modern stack tooling is not available unless I have a dedicated hosting package – VPS (virtual private server).

AI did not remove the need for platform knowledge in order to achieve what was required.

What I Deliberately Did not Do

Did not touch or write any of the SQL DDL scripts or create any of the selects or views used in the visualisation.
No backend coding
Minimal refactoring and clean-up – It was very difficult to just provide a basic visualisation without making it presentable for me, despite the intent of this exercise. I have a list of data storytelling and visual design principle articles in my list of future content. Surfacing a vanilla chart with random colours made me want to cry. I have separated that effort and credit consumption above.

The intent was not about perfection, it was about how quickly can I get to a working product.
This shift, I expect, would not be natural for most system thinkers and architect minds, I stared at the opening dialog in Kiro for a while and could not make myself click VIBE, I clicked SPEC he he he.

*Kiro’s Let’s Build screen, selecting between Vibe mode and Spec mode.*

Key Takeaway

Kiro didn’t replace development, it lowered the friction. It compressed the path from idea to working system. Even writing a technical blog article takes me longer to construct than the time I spent with Kiro. The biggest shift was not technical it was cognitive.
The bottleneck moved from code generation to defining intent and context clearly (you see what I did there).

What’s Next

I stopped myself from continuing to refine the visualisation (intentionally). I will continue to do this and discuss the refinement from a data storytelling, visual design and user experience perspective.
This was just about delivery speed and feasibility.

I’ve spent years building end to end data platforms the traditional way, this was different. Not because it was “easier” – though demonstrably it was; but because it changed where the effort is going and how I needed to work.