From what company size/number of developers/frequency of deployments/number of users do you think an an ops guy/team should start implementing the principles of platform engineering, building their IDP? How much is too much at the start?

I think it depends, a small team <100 could use something Humanitec/kubefirst/qovery to get going fairly quick

Totally agree with the laying the foundation plan. IME taking freedoms away in preference for secure/standard processes is extremely painful. So if you can build a muscle memory early on with engineers on how to request on demand resources, where to look for support, and things like that you will be in a much better place. Even if it's all just a mechanical turk style implementation behind the scenes!

Gotcha, thank you for your answers.

If you’re small enough to live with an opinionated framework / PaaS like Heroku, then you should do just that to be up and running as quickly as possible. Then observe diligently if you either outgrow the technical capabilities or start to notice communication and alignment issues in the org which follow along the lines of your architecture. The main thing here is - either you build your platform or it will build itself. There’s no way to not have one.

x2 and there are lots of solutions out there for smaller orgs looking to go external, get going fast, and avoid initial IDP hiccups. Where I work, Divio, is one of these external platform products. Even if I didn't work there, I would agree with jay that going with something pre-fab makes good business sense until you find product-market fit and understand what you need more deeply. A growing org can look for an opinionated option that matches their needs, without having to go to full Enterprise pricing.

If it's useful I shared my experiences/opinions and concrete demo just yesterday and the video is up here: https://www.youtube.com/live/AUyq-Ns5dUo (starts around 58min into the feed)