suchintan

Shoot - good call. For anyone reading this: Playwright is a very popular browser automation library.

Developers use it to automate websites, spanning use-cases like QA Testing, Web scraping, Form filling, etc

https://github.com/microsoft/playwright

We did the same thing. It took me a few iterations to get the library right, but overall I used approx 10 billion tokens here across a codex and Claude max plan

The first iteration was a rewrite using codex /goal but it ended up producing junk

The second iteration was fable controlling gpt-5.5 agents to write the code with really strong acceptance tests. This pushed the library in the right direction and is basically what you see here today

That makes a lot of sense. Sometimes it's easier to leave the baggage behind. It's too bad..selenium is a masterpiece. Thanks for sharing it with the world

This makes sense. I guess I wanted to understand why starting from scratch was better than "fixing" selenium, but perhaps "fixing" selenium isn't an option?

This is very cool. We were thinking about doing something very similar with Skyvern

What was the reason you went down this path instead of extending selenium with AI features?

maybe https://github.com/Skyvern-AI/Skyvern ?

You're right, but this is where the LLMs are especially useful. Our customers all prompt it to terminate if it doesn't have the right information / the pre submission confirmation doesn't match

What are some of the risks? This is a public web form available on the IRS website

Send me an email suchintan@skyvern.com - we can get you started

No, we don't have a lot of usage in that direction. People mainly use us to log into websites and either fill out forms or download files!

Unrelated, but thoughtful gave us some very very helpful feedback early in our journey. We are big fans!

That's the dream

Definitely. What are your thoughts on the CloudFlare agent identity

It's funny, one time we had a customer that wanted to use us to test their website for bugs..

Skyvern kept suggesting improvements unrelated to the issue they were testing for

We do have them! We are HIPAA compliant, have soc-2 type 2 and offer self hosted deployments

This is really cool. We might integrate this into Skyvern actually - we've been looking for a faster HTML extraction engine

Thanks for sharing!

I have a 2yo and it's been surreal watching her learn the world. It deeply resembles how LLMs learn and think. Crazy

Yeah, reverse engineering APIs is another fantastic approach. They aren't enough if you are dealing with wizards (eg typeform), but they can work really well

IF you can use crawlers, definitely do.

They aren't enough for anything that's login-protected, or requires interacting with wizards (eg JS, downloading files, etc)

I think they're complementary, and that's the direction we're headed.

We can ask the vision based models to output why they are doing what they are doing, and fallback to code-based approaches for subsequent runs

We can do it remotely with Skyvern if you're interested

Have you given Skyvern a try? Integrating it with n8n might do the trick

Have you ever given Skyvern (https://github.com/Skyvern-AI/skyvern) a try? I'd love to hear your opinion

That would be amazing!!

Very cool. The benchmark can be found here if you want to take a look at it: https://github.com/Halluminate/WebBench

We definitely plan to expand it. I want to get to ~10,000 for a reasonable benchmark.

15 blew my mind -- it's too easy to overfit that dataset

Definitely. It gives you a lot more control on which models you want powering these browser agents, which is the important part

Hey HN, we were playing around with MCPs over the weekend and thought it would be cool to build an MCP that lets Claude / Cursor / Windsurf control your browser: https://github.com/Skyvern-AI/skyvern/tree/main/integrations...

Just for context, we’re building Skyvern, an open source AI Agent that can control and interact with browsers using prompts, similar to OpenAI’s Operator.

The MCP Server can:

- This allows Claude to navigate to docs websites / stack overflow and look up information like the top posts on hackernews - https://github.com/Skyvern-AI/skyvern/tree/main/integrations...

- This allows Cursor to apply for jobs / fill out contact forms / login + download files / etc - https://github.com/Skyvern-AI/skyvern/tree/main/integrations...

- Connect Windsruf to take over your chrome while running Skyvern in “local” mode - https://github.com/Skyvern-AI/skyvern/tree/main/integrations...

We built this mostly for fun, but can see this being integrated into AI agents to give them custom access to browsers and execute complex tasks like booking appointments, downloading your electricity statements, looking up freight shipment information, etc

I've seen people use semantic versioning to version APIs and SDKs

https://semver.org/

We at Skyvern are still doing patch versions only

I wonder if more companies should open source their eval model outputs alongside the eval results

We tried doing that here at Skyvern (eval.skyvern.com)