I came across a post from an old colleague earlier today about a new AI coding tool from AWS called Kiro. Perfect timing, really – I’ve been keen to build another app that I felt was missing during my recent trip to Europe with mates.
We struggled to keep track of expenses as different people paid for various parts of our trip, and we tried to manage it all in a spreadsheet. I was thinking there should be an app that handles all this automatically, something like Splitwise.
Now, why would I build something if it already exists? Well, Splitwise Pro costs $4.99 USD per month, and I thought: why not build the bare bones of what I need as an experience with Kiro? Especially if it doesn’t cost me anything during the trial phase.
First Impressions: More Like a Co-pilot Than a Cloud Agent
After downloading and installing Kiro, the first difference I noticed compared to my experiences with Replit and Hostinger’s Horizons platforms is that Kiro is more like a co-pilot agent rather than a cloud-based AI coding agent. On first look, it seems a lot more powerful – it promised to build out your specs, ensure test frameworks exist, so you aren’t exposed as the agent works around your codebase. This is an area I needed to work on while building my NextlyGTD app. I’ve got an article in the backlog to discuss this in detail.
The other thing that jumped out at me was the platform’s ability to build an app that could be deployed to both web and mobile, which wasn’t easy out of the box with Replit.
The Setup: Local Development with All the Trimmings
After getting started, I immediately noticed that this wasn’t a cloud-based agent like both Replit and Hostinger Horizons. I needed to install a whole bunch of tools onto my computer to get started with my app. Kiro is an agent that works with the development tools on your machine, so you need to install all the frameworks you plan to use in your app so that Kiro can work with them.
This immediately reminded me of one of the drawbacks I found with all the copilot tools that existed before I discovered Replit. There’s a barrier to entry here that might not be a barrier for professional software developers, who likely already have most of the foundational libraries and tools installed to build apps. But for a non-technical person, this can be intimidating.
There seem to be lots of safety controls in place, which prevent Kiro from going crazy on your computer. It constantly asks your permission to proceed when it needs to interact with sensitive parts of your application architecture or computer. Again, for a professional software engineer, this would be comforting. But for a non-technical person, it’s unclear whether allowing permission in one instance over another is dangerous or not, so you pretty much default to just giving permission so you can move forward. This very much fits into the Terms and Conditions checkbox paradigm – who actually reads the terms and conditions anyway?!
Building the App: Claude 4.0 Under the Hood
After methodically installing all the libraries we needed to build the app, we finally got to building the app itself. The underlying agent is Claude 4.0, but you can also select Claude 3.7 – no idea why anyone would choose to do that. Maybe using 3.7 will be a bit faster or cheaper in the professional versions, or perhaps there’s the intention to add other agents so you can select which agent to work with on different types of problems. Again, a professional software engineer probably wants this granular control.
To build the base version of my app, I instructed Claude on what I wanted to build and specified that I wanted a dual web/mobile architecture. We went back and forth a bit on some architectural decisions, and we were ready to begin.
Where Things Started to Fall Over
This is where things started to fall over a little. The moment I created a folder/workspace, a new context window opened and all the work I had previously done with Kiro was lost. This might have been user error, and if it was, apologies to Kiro, but I ended up having to respecify what I wanted. This time, Kiro was a lot less patient and started building before we had clarified exactly what we were building. This meant we had a couple of false starts.
My advice to AWS: you need to work on the UX here.
After installing Expo to enable dual web/mobile development and testing, Kiro built the first version of the app for me. It took quite a bit of tinkering to get the web app up and running on my localhost, and I couldn’t quite get to the point where I got the mobile app working via Expo.
I noticed that Kiro would often just go missing in our interactions. The IDE would say that Kiro was working, but there didn’t appear to be any activity really going on. Potentially another UX problem here. I spent more time on environment configuration and troubleshooting than I would have liked – well, much more than I did with Replit when I built the first version of NextlyGTD.
The Reality Check
I only spent a couple of hours or so before my family called on me to stop. We’re on holiday, and I was just playing around on the couch with my laptop after my morning run and yoga session (as you do) while I waited for them to get up and get ready for the beach.
My General First Impressions
I think that AWS Kiro has lots of potential, especially for professional software engineers in corporate environments. It adds a level of professionalism and a more robust end-to-end development process. It lets you keep all your IP in your local environments, which I imagine would be a large concern for most corporates. I think it’s more suited at the moment to software engineers or highly technical stakeholders than it is to non-technical people, just given my initial impression.
Who Should Use What?
If I was deploying AI coding agents within my organisation: I would immediately start a project to assess Kiro within my organisation.
If I was a slightly technical person looking to build a new application: I would seriously explore Kiro as my coding agent because it offers much more flexibility to build multi-platform apps, and potentially to onboard additional engineers in the future.
If I was a non-technical person, and I could get away with a web-first experience: I would just stick to a platform like Hostinger Horizons or Replit.
If I was a non-technical person, and my app was mobile-first: Then it would be a tough decision between Replit and Kiro. Replit has a process to deploy from web to mobile, but it isn’t tightly integrated into the development process, which would make future releases complicated.
An alternative might be to start with Replit, build out the full web version, and then export the code and start using Kiro to move to a multi-platform development approach.
Final Thoughts
It’s awesome to see all these new tools coming to market. I’m very excited about these developments and the opportunities they offer for innovation. The AI coding space is moving fast, and tools like Kiro are pushing the boundaries of what’s possible for developers at all levels.
Have you tried AWS Kiro or other AI coding agents? I’d love to hear about your experiences in the comments below.
Leave a Reply