A.I. driven Smoke Testing

We can all agree that smoke testing our applications in production is valuable. Tests that fail to complete key interactions is often the first line of defense against regressions and give Engineering teams an edge to minimize or prevent outages that cost real businesses real money.

The lovely live Sketch by Johannes at the Automation Advocates Meetup

However, the issue with these tests that rely solely interacting with the frontend of an web application are susceptible to breaking to rigidity in HTML structures, A/B experiments, or even small changes like pop ups blocking elements from being clickable.

We go over two new techniques possible through leveraging A.I. to solve these problems.

Watch the full talk below, and read about each technique along with the sample workflow that you can copy to your Pipedream account.

Slides from the talk here.


Finding selectors over HTML

The first technique is to leverage the new LLM models by OpenAI that accept very large prompts.

Instead of maintaining ridgid selectors directly in UI tests code, instead leverage prompting to query the entire HTML document for specific selectors.

Opening this link below will copy the sample workflow to your Pipedream account. It already includes the steps to scrape a website and pass the HTML and prompt to OpenAI:

https://pipedream.com/new?h=tch_qKf1Yl

Screenshot driven assertions

The second technique is to use screenshots of actual webpages to perform acceptance testing and assert elements are visible and clickable.

Opening this link below will copy the sample workflow to your Pipedream account. It already includes the steps to scrape a website and pass the HTML and prompt to OpenAI:

https://pipedream.com/new?h=tch_knfrg0

Learn more