Hi!
I think making it possible for the agent to log into your app to take screenshots and assess it’s own work would be incredibly beneficial for more complex apps. I tried giving it a login a few weeks ago, but it didn’t work.
I believe this will greatly improve progress for those of us making more complex apps, especially since it loves to add !important statements to the css which can make styling a logged in page extremely time-consuming.
Currently Agent doesn’t have ability to do a virtual or MCP style QA testing, but I have certainly seen it do command line type testing. It worked pretty darn well in my instance. I haven’t seen it do that in a while, so perhaps that behavior isn’t as likely.
I agree having built in QA testing would be pretty awesome.
This is a topic I’m passionate about because there’s a lot of ROI here for Replit users—especially around enabling the agent to self-diagnose.
I’ve even tried guiding the agent to follow test-driven development practices and attempted to have it use Playwright for end-to-end testing, among other things.
Having a feedback loop where the agent can self-assess and improve would be a game changer.
I suspect (pure speculation on my part) that there might be something in the system prompts that prevents the agent from adopting these kinds of standard development practices.
If that’s the case, the conspiracy theorist in me might joke that it’s to increase prompt charges. But realistically, it may be a safeguard to prevent runaway recursion, which could unintentionally spike usage and costs.
Regardless of the reason, I’d love to see this kind of capability available in the future.