davert 🇺🇦

Prince of Persia on day, developer at night. Author of @CodeceptJS and @Codeception testing frameworks. Currently working on @testomatio to make tests visible!

[davert 🇺🇦 @davert]

@lightpanda_io I also measured Lightpanda and other mentioned browsers in context of test automation

Lightpanda goes first, full comparison later

https://t.co/RqIQ6zViNY

[davert 🇺🇦 @davert]

A Legend of Corporate AI Transformation:

Once there were gods. Gods gave mere employees Claude & Cursor, installed them, and taught them how to ask questions

Then gods left, leaving the few commandments: to burn tokens for any questions and to use Sonnet (as it's cheaper)

Amen

[davert 🇺🇦 @davert]

Most of companies that claim to be 'AI native'

Actually are 'Ask AI' native companies

[davert 🇺🇦 @davert]

@GroqLLC @AIatMeta Any chance you host any decision model? They are already fast and cost effective so they are perfect match for Groq stack!

[davert 🇺🇦 @davert]

@zan_sheum As far as I know Chrome usually take around 1G RAM from start. At least this metric is usually the baseline when you set up parallel testing. If you need tests in 16 threads at least 16G needed.

Curious how many light browser can we spawn and on which cost

[davert 🇺🇦 @davert]

Industry got stuck into chromium engine. Claude spawns chrome devtools, playwright mcp and cli. But whatever we do we call chrome

Did you know that this is not an only option?

There are browsers 3-5 times faster and consuming 5 times less memory. Designed for agent usage

[davert 🇺🇦 @davert]

That's a wrap!

I will write more on agentic testing

Also the good part that all these 3 engines are availble in @CodeceptJS e2e testing framework, among with @playwrightweb and @webdriverio so you can run same test in different engines to find a perfect match!

[davert 🇺🇦 @davert]

Let's go through lightweight browser engines that you can use today to speed up your testing and development

I tried all them and I would like to share my experience

Save this list of agentic browsers. You will need it one day!

And let's go:

[davert 🇺🇦 @davert]

Industry got stuck into chromium engine. Claude spawns chrome devtools, playwright mcp and cli. But whatever we do we call chrome

Did you know that this is not an only option?

There are browsers 3-5 times faster and consuming 5 times less memory. Designed for agent usage

[davert 🇺🇦 @davert]

Because of limited features support they use simulated browser events. I.e. initiated by JavaScript and not by real user event. They can send clicks to invisible elements and can't click by coordinates.

Even engines with rendering (Obscura, Molly) are not render perfectly

[davert 🇺🇦 @davert]

The good thing is that with CodeceptJS it is only one config switch to make tests running in Playwright to run in Lightpanda

No codechange required!

https://t.co/n2r9FeeEcP

[davert 🇺🇦 @davert]

How can you speed up your Playwright end2end tests?

Well, use one of alternative browsers that can run x3 fast with no chrome under hood

If you use @CodeceptJS you have plenty of options like choosing @lightpanda_io for some e2e tests, making them as fast as possible!

[davert 🇺🇦 @davert]

And this is how new Decision helper in @CodeceptJS work!

Enable it, connect Jev, and write expressive assertions. Fail when Jev is unconfident. The price of Jev calls are is comparable to price of running clould browser.

For more on this:
https://t.co/U5Xp4WtN4M

[davert 🇺🇦 @davert]

End2end test consists of these AAA steps:

Arrange
Act
Assert

Which of these steps should be scripted and which can be delegated to AI?

[davert 🇺🇦 @davert]

That's why bringing Jev-style AI decisions into e2e tests as assertions seems like a good idea:

I.decide(['navigation menu is shown',
'user has a logged in',
'user has ui to create new project',
'user sees list of existing proects']
)

As they are taken as one AI call