Software automation musings

Today, I had to automate the migration of inpractise.com's video provider from Contentful to Vimeo. The process implied downloading every video (around 500) and re-uploading it to Vimeo. Then linking the video to the corresponding article.

We thought of hiring someone to do this task, but as an engineer, I decided to write a small script to do this task instead. This is a tedious task that would be error prone, and subcontracting it would not be scalable.

Every time I write these kind of automation scripts, it feels like I am teaching a robot to do something I want to do. It is a good exercise to realise the actual power of software engineering.

There is one bottleneck though: "Can I type the rules for the task?". Not all the problems actually fit this description, but some like migrating to Vimeo does look like a problem I could define.

I think there are other kinds of problems in which you can tell if the result is correct or not, but you can't really explain how to do the task. I believe that would be one of the domains for machine learning (i.e. image recognition).

And I'm sure there are problems in which you can't tell if the outcome is good or not, and you can't explain how to do the task either. What technology would allow us to tackle these problems? Is it reinforcement learning? Are humans still required for those problems?

Just a quick random reflection on software automation.