Really good points. On quality it’s not replacing human insight, but it is exceptional at pattern recognition and coverage at scale. It catches edge cases that tend to get missed and never forgets past regressions. The best results I’ve seen come from pairing the agent with human QA. The agent does ambient monitoring and flags suspicious behavior. Humans then dig deeper.
Cost-wise, it’s surprisingly reasonable. The version I saw ran in containers that spun up based on commit activity or deploy frequency. So if no one is pushing code, it's idle. But during launches or busy dev cycles, it ramps up. Much cheaper than staffing a full team to maintain 24/7 vigilance.