I have several digital employees running automations via Cloud VM Agents. However, they do not consistently load/capture the native Browser UI/UX verification experience. I frequently catch my agents in post-run analysis installing Playwright/Playwright MCP in order to faithfully follow UX verification instructions.
Specifically Codex doesn’t appear to “ever” find the native CUA, Sonnet finds it pretty consistently, and Composer is able to find it. Is there some recommended setup I should be using for this? I theoretically would also like to target my automations to the LVP model for the role of the employee, though my automations seem to all collapse to a single model that I select. I’m fairly new to this whole abstraction, but am currently focused on increasing my meta-layer orchestration beyond menial tasks like “do your job” but end up spending a fair amount of time in Agent verification. Any tips?
Ah - interesting. I’m looking at my Automations run history and you are right - I think I ended up re-running some features via “Cloud Agents” manually when they failed via my Desktop Cursor (Agents Window). Thanks for the info! I can figure a way to get Playwright working OOB for the Automations for now - is there a release date planned for Automations CUA access?
Actually, I realized I can just use OpenClaw for VM hydration & ITTT right? Since the SaaS launch is more consistent about the snapshot used (and I made some proxy logic to ensure it’s using the right hydration) - that should effectively allow for “VM CUA” via automation.