Brendan
Gregg

There is a good chance that the performance tool your team uses today - the flame graph you open when something is slow, the eBPF tracer that tells you what the kernel is actually doing at 3 AM - came from one person working in Australia, obsessively dissatisfied with the idea that computers should be opaque. That person is Brendan Gregg.

Right now, Gregg is a Member of Technical Staff at OpenAI, working remotely from Sydney on the performance of ChatGPT's datacenters. Before that he was an Intel Fellow, the highest technical recognition Intel gives. Before that he spent nearly a decade as the performance engineering leader at Netflix. The pattern is clear: wherever the world's most demanding compute workloads are, Gregg shows up and makes them cheaper to run.

His career has a quality that is rare in engineering - every job he has taken has been at the exact intersection of "unsolved and expensive." At Sun Microsystems in the mid-2000s, he built ZFS L2ARC, a flash memory caching layer that gave file systems a new tier to exploit. At Joyent he mastered DTrace, creating the DTraceToolkit that became the standard reference for dynamic tracing. At Netflix he invented flame graphs because he needed to understand a CPU performance mystery and the existing visualizations weren't good enough. He built the tool he needed and then gave it to the world.

That combination - deep system expertise, practical invention, and reflexive generosity with knowledge - is what makes Gregg unusual. He does not collect insights the way some engineers hoard competitive advantage. He writes them down in 400-page books, puts the source code on GitHub, and teaches workshops until 2,000 engineers understand something they didn't before.

Ask a performance engineer what Brendan Gregg is known for and they will say "flame graphs" before you finish the question. That answer undersells it. Flame graphs did not just help engineers find slow code. They changed the visual language of performance analysis for an entire industry.

The idea is deceptively simple: take a stack trace, aggregate all the samples, and draw them as a stack of horizontal bars where width means time spent. The resulting picture looks like fire - hence the name. But the insight is that you can see the entire execution profile of a program at a glance. The widest bars at the top are where time is being lost. No guesswork required.

Gregg invented the visualization while working at Netflix to solve a real mystery: the servers were burning CPU and nobody knew exactly why. After building the first flame graph, the answer was immediately visible. The tool was open-sourced, the paper landed in Communications of the ACM, and within years every major tech company - Google, Facebook, Microsoft, Cloudflare - had adopted it. It is now a default feature in most profilers.

At Intel he extended the concept to AI hardware, creating AI Flame Graphs that visualize GPU and xPU performance stacks all the way from the hardware up through software. Now at OpenAI, those tools are pointed at some of the most expensive compute infrastructure on the planet.

Gregg grew up in Newcastle, New South Wales, earned his degree from the University of Newcastle, and started his career where most engineers do not - as an independent consultant and trainer. He taught performance tuning and system administration to over 2,000 engineers before he was famous for anything. That pedagogical instinct never left him.

Sun Microsystems gave him his first serious stage: kernel engineering, ZFS development, DTrace mastery. When Oracle swallowed Sun, he moved to Joyent where he built the DTraceToolkit and cemented his reputation as the person to call when a system was misbehaving in ways no one else could explain. He holds two U.S. patents from that era - one for intrusion detection systems, one for data caching methods.

Netflix was the chapter that made him a household name in engineering circles. From 2014 onward, he designed, evaluated, and tuned performance across one of the world's largest streaming infrastructures. The flame graph was born there. The USE Method was codified there. The books - the real books, 1,000+ pages of dense, practical knowledge - were written during those years.

Intel came next, with the title of Intel Fellow - a distinction held by only a handful of engineers at any given time. He applied the same instincts to AI hardware: xPUs, GPUs, IPUs. He built AI Flame Graphs. He wrote about what he was learning, as always.

Then OpenAI called. He said yes because the cost of AI compute is genuinely staggering - environmentally, economically - and because his hairstylist was using ChatGPT every day and knew what it was better than most people knew Intel. That combination of civilizational stakes and real-world adoption was enough.

Gregg writes books the way he builds tools - because something needs to exist and doesn't yet. His four major titles are not introductions. They are field manuals, the kind you dog-ear and keep near the terminal for years.

In early 2026, Gregg published a blog post titled "Why I Joined OpenAI." It reads unlike most corporate announcements. He writes about his hairstylist Mia and how her casual enthusiasm for ChatGPT landed harder than any pitch deck could. He writes about the scale of the problem - AI datacenters consuming staggering amounts of power, with costs growing faster than efficiency improvements. He writes about how OpenAI's engineering culture felt familiar: the same urgency, the same freedom to change anything, the same pressure of scale he knew from Netflix.

Greg Brockman, OpenAI's President, greeted him personally and described being a longtime fan. That is the quiet version of how well-regarded Gregg is in the industry - the president of one of the world's most prominent AI companies was excited that the person who invented flame graphs was joining the team.

His role is titled Member of Technical Staff, reporting to Justin Becker. The work is performance engineering for ChatGPT's infrastructure - the same discipline he has practiced for two decades, applied to workloads that did not exist when he invented the tools to analyze them. He works remotely from Sydney, where he has been based since 2021.

The childhood dream closes a loop here, too. Gregg grew up watching Blake's 7, a British science fiction series whose centerpiece was Orac - a supercomputer of extraordinary intelligence. He wanted that future. Now he is writing code that helps run the closest thing to it.