Brendan Gregg on being copied as an 'AI Brendan' (www.brendangregg.com)

Performance engineer Brendan Gregg reports multiple AI performance‑engineering agents have been built using his public work—some helper agents that interpret flame graphs and eBPF metrics (nicknamed "AI Brendan") and some “Virtual Brendan” products trained on his talks, blogs, tools and book pages that claim to replicate his tuning expertise. He stresses these tools are partial: a trained "virtual" can automate roughly 15% of what he does, will age quickly as tunables and practices change, and is built on an incomplete snapshot of his private experience. Gregg also notes he isn’t involved with these products and has long used AI tools himself for research and coding tasks. Technically and commercially this raises hard tradeoffs. Many agents use pattern‑matching on flame graphs/metrics to surface fixes and can yield ~10–50% gains for teams lacking performance engineers, but ROI is hard to quantify. Pricing models (e.g., $20/instance/month) are tricky because customers can apply discovered fixes fleetwide, and secret auto‑tuners violate change control and risk being blamed for outages. There are also engineering burdens—orchestration, UI, logging, cross‑runtime support—and legal/copyright questions about training on published material. Gregg’s verdict: these agents can be useful (especially in‑house or open source), but they won’t replace deep human expertise and may erode the industry’s “performance IQ” unless responsibly designed and upstreamed.