If you think about "price", "speed" and "accuracy (reliability/quality)", our bet is that models won't hit those 3 together. So you won't get a model that is very fast, very cheap and very accurate anytime soon.
Also, imagine that you have a case where you want to scrape 10,000 records from a website, why have AI navigate to every page to do this? why not write the code, run it, and get consistent and fast result? its also predictable, if it messes up, you know what happened and you can trace it to the exact line of code.
I’d like to see less focus on Playwright and more focus on giving the agent more than just an MCP to browser automation. Make it multi-modal, figure out how to optimize when to send screenshots to which model, etc… current coding harnesses are awful at any UI automation because they’re just automating DevTools and occasionally screenshotting. It’s obviously robotic, it’s slow, it’s ineffective and makes it difficult for the agent to validate success of code changes.
Generalized computer use is what will ultimately solve this, but I think there’s real intermediate value in optimizing browser workflows specifically, as a medley of remote browser automation and multi-modal browser use.
our product is infra + agent. You can use codex or other agent to generate the code. We actually have a cli that allows you to deploy projects to our infrastructure.
We are actually working on open sourcing a plugin that you can use with any coding harness!
Also, imagine that you have a case where you want to scrape 10,000 records from a website, why have AI navigate to every page to do this? why not write the code, run it, and get consistent and fast result? its also predictable, if it messes up, you know what happened and you can trace it to the exact line of code.