An AI Agent Did My 25-Minute Task in 75 Seconds. So I Made a Free Course About It.

"Am I being replaced?"

I've worked in performance engineering for twenty-five years, and I've never heard that question asked as often as in the last two. Engineers ask it in Slack DMs, at conference bars, in the quiet moment after a demo. Few ask it in front of their managers.

I decided to answer it with measurements instead of opinions. Then I turned the answer into a free course. This article is the short version.

The experiment

Correlation is the least loved job in performance testing: dig through responses, find the dynamic values, write the extractors, wire them up, hope you caught them all. It sits at the centre of the manual grind that eats somewhere near half of a test week.

I timed two approaches on the same task, same application, same HAR file captured from browser dev tools. No APM, no special tooling.

Doing it by hand, with ChatGPT assisting, took me 25 minutes and 20 seconds. An AI agent completed the same task in 75 seconds. That is 20.3 times faster, and it reclaims about 24 minutes of engineer time on every single correlation.

Speed was the headline I expected. Accuracy was the one I didn't. The task contained six correlation candidates. The agent found all six. I found two. Manual work missed two thirds of them, and across the run I logged six human errors against the agent's zero.

Sit with that for a second, because I had to. A machine beat twenty-five years of experience at a core task of my profession, on quality as well as speed.

Why that isn't the whole story

Here is the part that took me longer to see. The correlation task was never where my experience lived. It was overhead: mechanical work standing between me and the questions I'm paid to answer. Does this system survive Black Friday? Why does p99 latency cliff at four thousand users? Which of these fifty findings matters to the business?

The agent removed the overhead. It did nothing to answer those questions. Judgment about risk, about what to test, about what a spike in a graph means for real customers on a real Tuesday: that work remains, and demand for it is growing.

How fast is it growing? Splunk's 2024 research put the cost of downtime for Global 2000 companies at four hundred billion dollars a year. That figure is the reliability gap, and it exists with today's systems. AI is now shipping code into production faster than humans can review it, which widens the gap further. The world needs more performance judgment, applied more often, than the current workflow allows.

That inversion is the core of the argument. A team that ran two full test cycles a week can run five. Twenty-four scenario runs a quarter become sixty. The backlog scenarios that never got tested, the payment timeout, the retry storm, the cache invalidation under load, finally get their turn. Somebody has to decide what those tests should be and what their results mean. That somebody looks a lot like you, minus the correlation work.

Who the course is for

I packaged all of this, the measurements, the workflow, the argument, into a free course called Intro to AI Performance Testing. Six videos, about forty minutes end to end. No code, no exercises, and no sales pitch inside the modules.

1. The Four Hundred Billion Dollar Reliability Gap: why demand for performance work is rising, not falling
2. AI Is Eating Performance Engineering: the experiment above, in full
3. Why That Should Excite You, Not Scare You: the career case
4. The New Performance Test Workflow: what the job looks like when agents do the grind
5. Real Results from Real Teams: measured outcomes, not vendor slides
6. Your Next Step: how to start from where you are

It's aimed at working performance engineers and test leads. If you've watched the AI wave arrive and wondered where you stand, the course was made for you. If you manage a performance team and need a shared starting point for the conversation, send them module one; it runs six and a half minutes.

Questions, disagreements, and war stories are welcome: find us on LinkedIn. The war stories are the best part.

Related: The 20x Claim, Audited · Why Performance Engineering Is the Future of Load Testing · Performance Testing vs Performance Engineering