Call-Cadence
An Adventure in AI Coding
The middle of 2025 I had an idea for a simple software tool:
CallCadence is a lightweight scheduler that uses HangFire to run and manage timed API calls. Configure static endpoints, payloads, and authentication once and let CallCadence keep your integrations on beat.
A simple enough programming exercise however, I wanted to challenge myself. So I decided to write it totally through AI, deployable as Docker Images, separate backend and UI. And, today I am drawing a line in the sand and releasing the first version:
Dashboard Example
This is hitting the Rest Test API I wrote about sometime back in order to generate data. The errors are purposeful.Docker Images (coming)
I’ll be detailing the journey over the next couple of weeks, but here is a high level overview.
The Good
This was actually better than I expected. I used GitHub Copilot, mostly in Agent mode. I did try creating a starter using GitHub’s prompt on a new repo. That was more complicated than I want to think about. It did create something but it was easier to start with a blank repository and craft it in stages than it was to take that starting point and make it what I wanted.
At $30 a month and using up that money in less than a week’s worth of coding, this took longer than it would have had I just written the code myself, but it was handy to pass the coding request off and then come back to it later to find it was close.
The Bad
The biggest problem with coding this way is that there are certain things that this type of coding cannot do. For instance, while the agent can run tests, it cannot use the software. The best example of this is when I requested a change, which worked perfectly, apart from the fact that I couldn ‘t log in. Any human programmer would have found this because they would have been attempting this while testing, at least I hope they would have.
The Ugly
I will have to go back through the private repository that contains all the work that led to this but I know that I had to spend tokens at least 3 times to address the same problem bcause the code change broke something that wasn’t caught by testing because it required actually using the software. Having a UI separate from the API backend and having authentication is not a radical concept, but it has delicate and fiddly bits where the cookies that handled authentication and authorization kept disappearing.
Not only that, the agent could not generate valid SSO code using my Authentik instance so that is going to be manually coded soon.
What’s Next
Now that I have published the repository, there are several more things I want to do:
Add GitHub Actions for publishing the docker images. I am currently using my personal docker repository for the dashboard instance.
Write an installation document with screen shots
Get SSO functioning.
And, write up more of the process with details.