Gojko is building up to the moment where he tells us that this is no longer
true:

"Record+Replay tools (like Selenium) are optimizing the time spent writing
tests. This is not where the value is."

Tools like the Galen framework enable you to test UI by describing expectations.
Still not optimal, since syntax like this is nothing your users would supply.

https://github.com/AppraiseQA/appraise/blob/master/README.md/ is the tool he
developed for visual approval testing. It helps to review and approve changes to
web pages, visual layouts and browser components quickly through visual
inspection using headless Chrome.

And with headless Chrome you can parallelize the tests using AWS lambda:
https://aws.amazon.com/blogs/devops/ui-testing-at-scale-with-aws-lambda/

"This (visual difference testing and running the browser used to test in
parallel) changes the economics of UI testing and makes them something we can
now start looking into, because tests are now easy to maintain and extend."

This will be really cool once we have a bot that adds screenshots to a PR when a
commit breaks the test, and reviewers can accept / reject changes to the test in
the PR. No need for a human to maintain the (small changes to the) tests
anymore.

Great talk @gojkoadzic 👏