I was watching overview of TempoLabs and my initial impression was “wow, that’s a lot of stuff” as they not only have AI generation but it seems they understand at least the basics of a software devlelopment process. When you invoke the AI generation, it firstly parrots back its understanding of what you’ve asked for. It then proceeds to generate mermaid diagram(s) and then on to the generation of wireframes. In this post we’ll take a look at TempoLabs to see how far we can get with our challenge of building an ABN lookup application.
As with the previous attempts, my input to the AI prompt is fairly simple: Create a cross platform application that can lookup the ABN of a business using an online service.
As the AI starts to get going I’m suitably impressed with the interpretation of my input and the subsequent mermaid diagram it generates. However, the experience rapidly when downhill from there as the wireframe generation phase seemed to break. I tried the usual reqeusts to the AI prompt for it to fix the errors but after multiple attempts it wasn’t getting very far. I even tried deleting the partially generated UI, only for it to again fail to generate the wireframes.
Ironically, as I was putting this post together, I returned to the app I was creating in TempoLabs and the screen must have refreshed because there was my generated wireframes.
Now that we had a UI, it needed to be connected to the ABN lookup service. After a couple of simple prompts, the AI had built a service that would then connect to the ABN lookup service. I was moderately surprised that it was indeed able to call the service. Interestingly, it didn’t quite understand my initial prompt which I was looking to retrieve the ABN for a business; instead it assumed I had an ABN that I needed to lookup. This was easily rectified by a prompt that clarified that I wanted the ABN for a business name that would be entered.
The result was a basic UI that would indeed lookup a business name.
The only point I would make is that during this process the AI realised that I needed to enter a API key in order to access the service. Instead of prompting me for the GUID and then putting it somewhere sensible (as other services like Lovable have done), it instead suggested I needed to create a .env file and add the GUID myself….. I promptly asked the AI to do it for me.
I then requested that the UI be changed to show all results. Somehow the AI generated code that caused an error, which it detected and then prompted me as to what to do.
I clicked the Fix with AI and the issue was then solved, resulting in a UI that did indeed return all the search results.
The weird thing about this exchange is that I’m not in control of the changes the AI is making, so when it makes a mistake and causes an error, it should be able to automatically iterate until it’s either stuck (unable to fix the issue) or it’s fixed the error. It would seem that prompting me with the error should be a last resort.
Ok at this point I’m going to give TempoLabs the win as I feel it’s been able to achieve what others haven’t – a basic application that’s able to retrieve ABN numbers for a business. Not withstanding the rough edges in places, this was a surprisingly good experience.
Could you continue building a real application with this tool? I think you could. However, you would probably have to sign up to one of their paid tiers because just this small set of experimentation used up over half (350K of 500K) of the daily tokens you get on the free tier.
Anyhow, here’s an update on the current scoreboard.
Human Developers: 2 AI Code Builders: 2
Also for reference here’s the current ordering of the services I’ve tried to date: