Building Apps with AI - AWS App Studio review

At some point in the future, AI will do almost all the heavy lifting required to build web applications. AWS App Studio appears to be Amazon's first step on this pathway. A preview was released in the summer of 2024. How good is it? Let's have a look.

App Studio has AI that works in two phases - firstly you specify the app you want in natural language. The model transforms this into 'requirements', which break down into three sections: Use cases, user flows, and data sources. It then summarises these into 'key features' and 'usage'. This summary section is really just AI sugar on top of the requirements, which are used to create the app.

The second phase is where the heavy lifting happens, the structured requirements are used to create an application. The first time I created an app with this was very much a 'wow' moment. I did indeed have a functioning application, complete with a relational database, a front-end with a side menu and various screens, and even some 'integration' flows for email notifications (which would need further configuration).

The second time I created an app was a bit less 'wow' - I got a very similar application - with a relational database, a front-end with a side menu and various screens, and some 'integration' flows. So it's clear that behind the scenes it's working in a very structured way with limited 'agentic' capabilities.

I decided to test it with something a little more challenging; an app for loan lifecycle management. I started with a basic paragraph outlining what I was looking for:

I want an app to manage the lifecycle of a loan. It should allow sales people to create loan applications, including amount, rates, fees (which should be chosen from a set list, managed by credit directors). The other stages of the loan are credit review, transaction execution and in-life monitoring. The monitoring team can mark loans as at risk using a traffic light categorisation. Users can subscribe to watch a loan, they will get email or slack notifications when loan statuses change.

The requirements it came out with looked reasonable, a good starting point perhaps. Interestingly, the requirements and overview are not editable - I needed to describe changes in the console. I asked the AI to make a couple of tweaks before clicking 'generate'.

This is where the magic happens. In around a minute, my app was created. It was a reasonable starting point but far from perfect - the fees management was not there, a couple of pages were mis-named, and basic data validation was missing (e.g. I could enter text for date fields). At this point, App Studio becomes no more impressive than other low-code application builders like ReTool and AirTable. There's an AI chatbot to answer my questions (and it did a good job with the few I had), but it can't make changes to the app. At least not yet.

It will be really interesting to see where AWS takes this over the coming months and years; this preview release is clearly not an MVP. There will be plenty of competition in this space (GitHub Spark for example), but the close connection to the AWS ecosystem could make App Studio a great option for companies already set up in AWS who want to build apps quickly and easily (who doesn't?!). AWS haven't shared their App Studio roadmap, but I'd love to see;

• More accurate conversion of requirements to apps

• The ability to change apps on the fly with natural language

• AI that can customise design as well as use cases

• Access to hundreds of connectors for services in and outside of the AWS ecosystem

• Change management linked to Github and AI to manage test environments, automated testing and deployment pipelines

How long will it be before this all becomes a reality? Time will tell, but for sure, in 5 or 10 years time the landscape will shift massively.