By Klexy from Aofeisi | QbitAI
Brendan Gregg, a top expert in system performance optimization, has officially announced joining OpenAI.
After joining, he will be part of the ChatGPT performance team, working remotely from Australia and reporting to team leader Justin Becker.
Brendan is revered in tech circles as the "God of Performance", and his arrival was personally welcomed by OpenAI President Brockman.
Brockman even stated that he has been a longtime fan of Brendan for years.
How impressive is Brendan?
His seminal work "Systems Performance" has long been required reading for performance engineering at universities and tech giants worldwide.
He also invented the famous Flame Graphs, which allow programmers to visually see what the CPU is busy with, just like looking at a heatmap.
He is also a major driving force behind the Linux kernel core technology eBPF, single-handedly building the modern toolbox for cloud computing performance analysis...
Netizens commented that Brendan's work is absolutely next level.
So why did such a tech leader choose OpenAI at this time? He explained his views and observations in a blog post.
Brendan's Blog: Why He Joined OpenAI
As soon as Brendan joined OpenAI, he set a strict rule for himself.
Do anything, do it at scale, and do it today.
This is a state of constant readiness for battle. In his view, in this industry, optimization must be fast—it must run on thousands of machines the same day and show immediate results.
This hard requirement for speed and scale forces him to constantly monitor those massive computing clusters. Even if he can squeeze out a tiny bit of efficiency from the gaps, he must act immediately without delay.
OpenAI's "no restricted zones" environment provides him with the space to experiment—at OpenAI, as long as performance can be improved, no area is considered too difficult to change.
Of course, beyond seeing OpenAI as a big stage, Brendan also discovered that the AI industry urgently needs his participation.
What triggered this realization was a real-life experience.
He found that hairdressers, real estate agents, tax accountants, part-time beekeepers, and people from various professions were all chatting with him about ChatGPT. This made him realize that AI has become a tool ordinary people use every day, and the traffic behind it must be enormous, with backend pressure having increased by an order of magnitude.
Facing such massive traffic, the old methods from the general-purpose computing era no longer work. Over the past few decades, people have been accustomed to focusing on CPU and database tuning, holding hammers from the old era. But now they face super-clusters built from tens of thousands of GPUs and complex neural networks.
The tools in the old toolbox can't fix the new machines. Facing this new species of large model training, one must set aside previous experience and develop a new set of engineering methods specifically for large models.
This is why he decisively left the general-purpose cloud computing field he had worked in for half his life and dove into the hard challenge of AI infrastructure.
His task is very specific: to solve the performance bottlenecks behind ChatGPT and ensure that this expensive machine runs blazingly fast even when people worldwide are using it.
Who is Brendan?
So what kind of legendary figure is Brendan?
He can be described as the "anchor" of modern system performance. His books "Systems Performance" and "BPF Performance Tools" are well-known in the backend and operations fields.
These two masterpieces are revered as bibles in the global tech community, specifically for solving the most thorny system bottlenecks. They are the standard reference answers and "last resort" when troubleshooting problems.
Brendan, who wrote these two masterpieces, is a veteran honed in extreme combat environments.
In his early years (2001-2014) at Sun Microsystems and Joyent, he was already a core developer of DTraceToolkit, establishing his status as a pioneer in dynamic tracing.
In the middle period (2014-2022), he moved to Netflix as a Senior Performance Architect.
At that time, Netflix was facing the world's largest cloud architecture challenges. He dealt with massive concurrent requests and extremely complex microservice architectures daily, handling performance puzzles that would never be encountered at ordinary scales.
In the recent period before joining OpenAI, he became an Intel Fellow.
In this top technical position at the hardware giant, he focused on solving a long-standing pain point—how to help software engineers understand the low-level data returned by hardware PMUs (Performance Monitoring Units).
Beyond his publications and career history, he is also an inventor of industry-wide analysis methods.
One of his most representative inventions is the Flame Graph mentioned at the beginning of this article.
Before this, analyzing CPU hotspots meant staring at thousands of lines of boring text stack traces with extremely low efficiency.
He transformed this data into intuitive, visualized interactive charts. Where there's a performance bottleneck, you can spot and eliminate it at a glance.
That wasn't enough. To compare performance differences before and after version updates, he derived Differential Flame Graphs.
Beyond monitoring what the CPU is busy with, he also promoted the Off-CPU Analysis methodology, specifically targeting the invisible killers that cause processes to "slack off" due to I/O waiting, completely filling blind spots in traditional analysis.
Additionally, the most standardized toolsets in the Linux ecosystem, bcc and bpftrace, are maintained and contributed to by him long-term.
He also developed Latency Heatmaps to reveal long-tail jitter hidden by averages, and the USE Method (Utilization, Saturation, Errors) specifically to guide confused troubleshooters.
In short, Brendan turned the "black art" of system diagnosis into a systematic science. The USENIX LISA Distinguished Achievement Award is the most authoritative recognition of his practical achievements over the years.
One More Thing
In his blog, Brendan mentioned that choosing to join OpenAI also fulfilled a personal dream.
As a child, he was a die-hard fan of the British TV series "Blake's 7" and was particularly fascinated by a supercomputer called Orac. Orac could control other computers in the universe but had a terrible temper and would often snap at people.
In college, Brendan wanted to build his own Orac, but the hardware at the time was too underpowered—the memory couldn't even store a complete dictionary. After being mocked by computer salespeople, the idea fizzled out.
Until he encountered ChatGPT. He found this thing was practically a living Orac.
He even specially modified ChatGPT's custom instructions to make it mimic Orac's haughty tone of "only primitives would ask such stupid questions" when conversing with him.
For him now, joining OpenAI is essentially continuing that unreachable childhood sci-fi dream with his own hands.
Reference link: https://www.brendangregg.com/blog/2026-02-07/why-i-joined-openai.html