You might have something here, AI agents that work on anti detect browsers is a good idea
the local api of browsers like adspower or dolphin is the foundation here . u spin up the profile programmatically , but the real trick is how u bridge the visual output to the agent without leaving a trace
we spent the last month testing a setup where the agent handles the proof of personhood steps on high-trust platforms . if u use standard playwright / selenium selectors , u carry a specific bot footprint cuz the interaction with the dom is too precise . an ai agent sees the pixels and performs movements with actual noise and biological-style latency
the technical hurdle u gotta solve
most sophisticated filters (like datadome or cloudflare's latest) are sniffing for the cdp (chrome devtools protocol) connection . if u leave the debugger port open while the agent is navigating , ur trust score tanks regardless of how human the mouse looks
we started using a detached bridge essentially the agent takes a screenshot , calculates the move , and we inject the input through a custom driver that doesnt keep an active debugger link . it adds a bit of latency but the survival rate for account farming on impossiblesites jumped by like 60% in our latest runs
if u combine this with a clean isp-level mobile proxy , ur basically invisible to current ml-based detection . it’s expensive to run at scale tho , so we only trigger the agent logic when a captcha or a behavioral trap pops up the rest is handled by low-cost scripts to save on the token burn