Hey YC!

We’re Chris and Jorrie, cofounders of BrowserBook.

https://youtu.be/FkaGHhJWy1w

TLDR

We make it trivial for developers to write reliable web automations.

Ok, but what is it?

BrowserBook is an IDE that puts all the tools automators need in one place, so the iteration loop to write, test, debug, and deploy web automations is significantly reduced.  It combines:

• A Jupyter-like coding environment where developers can run code cells individually,
• An inline browser to run automation code against
• A Playwright coding agent that can write the automation for you
• A bunch of built-ins for taking screenshots, extracting data, and authentication

And, once you’re done writing your script, you can deploy it and trigger it via our public API.  It’s super easy!

But browser agents are a thing, right?

Yes, and we’re not using them.  We built BrowserBook because of the shortfalls we experienced with browser agents and the pain of actually writing reliable scripts.  At a very high level, in the AI world you can take two approaches to web automation - write scripts, or use browser agents.

1. Use browser agents: ostensibly more flexible, but expensive, slow, and unreliable
2. Write scripts: fast and debuggable, but infamously brittle

Which leads us to…

The problem (and how we got here)

Sometimes we need machines to interact with web pages, which are made for humans.

This is basically the whole challenge.  There’s a lot of reasons we might want machines to poke around on the web:

• Agents using a browser to interact with the web to perform actions like humans, especially when they can’t use APIs
• Scrapers to gather data that’s useful for our business
• UI tests that can interact with a website the way a human would to ensure everything’s working as intended
• Workflow automation via RPA

We tried browser agents in our pre-pivot days to automate healthcare workflows. They mostly didn’t work; they were an order of magnitude more expensive than scripts, slow because of all the context they needed, and tended to drift when workflows got complex.  When they did drift, fixing them was just tedious prompt-massage and trial-and-error.

We ended up giving the agent a library of scripts to execute, and came to realize scripting was simply the better solution for reliable enterprise workflows; they’re deterministic and therefore inherently debuggable, not to mention their cost and speed advantage.  They with a host of issues though:

• Websites change often causing scripts to break. We needed a way to quickly diagnose and iterate on our scripts.
• Setting up browser and filesystem infrastructure is not easy but very necessary in the web automation world.
• It’s just annoying to write scripts - if you’ve ever had to dig around for selectors, you know: it’s tedious and time consuming.

So we built BrowserBook to solve those issues.

How does it work?

1. Run code piecemeal - if you’re fixing a breakage 2/3rds of the way through your script, you shouldn’t have to run the whole thing every time you test it.  We took inspiration from Jupyter notebooks and allow you to run code cells independently, while maintaining context and state:

   

2. No more digging for selectors - tell the coding assistant what to do and it’ll implement the automation logic in Playwright:

   

3. Built-ins for the hard stuff - want to take a screenshot?  Log in to a website with 2FA?  Extract data?  We’ve got built-ins for those, ready to go out-of-the-box.

   

4. Accessible via API - if you need to execute your automation via an agent you’re building, scheduled via a cron job, or as a part of your application, you can trigger automations remotely and watch the playback in-app or via our artifacts API endpoint.

   

Our ask

Try it out!  If you:

• Run browser automations
• Are building agents that need to interact with the web
• Need to write a one-off automation that will make your life easier

BrowserBook is available for download for Mac users on our website, and if you’d like you can book a demo here.

Additionally, if you know individuals, startups, or teams at larger companies working on browser automations, we’d love to meet them!