Here's the funny part. I've spent 15 years in QA — manual testing, then automation, then test development, then test management. I've held just about every title this field has. And on day one, at a new desk, the feeling was as basic as it gets: I have no idea what I'm doing.
The way it happened was mostly luck. After my last company laid me off, I didn't really look for work. I spent the time on dev.to, writing my 36 Stratagems series — 30 down, 6 to go. That stretch was honestly pretty good.
Then an old colleague called. He was hiring testers and asked if I knew anyone. I thought about it and said: how about me.
Why I said yes is simple. It's an AI agent startup, a year or two old, sitting right on the hottest thing in the industry — and in a vertical, not another general-purpose wrapper. That was enough on its own. Poking at agents and models is what I do for fun anyway.
The first is process. The product → requirements → dev → test → ship pipeline here is genuinely different from what I'm used to. I'll be honest: the first week I thought it was a mess, and I'm still adjusting. When you're used to one kind of order, a different kind just looks wrong. Usually it isn't wrong. It's different.
The second is pace, and this one's good. My last company ran on instant replies. Requirements changed, and the PM was expected to answer on the spot. Here it feels slower, in the best way. Testers get more time and more room to make their own calls. You're allowed to think something through instead of getting dragged along by it. Two weeks in, this is my favorite thing about the place.
Do QA long enough and you learn something: slow is fast, and fast is slow. Time you steal up front gets paid back with interest, in rework and firefighting. Half a day spent on the edge cases, the weird inputs, the what-ifs — that half day usually buys back a week. So when I say this place feels slow, I don't mean nobody's in a hurry. I mean the pace goes where it counts.
Fifteen years buys you fast pattern recognition. You look at something and the problem jumps out at you. It also buys you a bad habit: when something doesn't match how you'd do it, the first instinct is to doubt it instead of figuring out why it's that way.
So my rule for these two weeks: understand first, judge second. A beginner's mind sounds easy. In practice it's an argument with yourself.
Mark was the reliable one — years of experience, all of it in his head. His company slowly extracted that experience into a skill. Then they laid him off.
I wrote that one from the outside, as the author, feeling pretty pleased with myself.
In that story, the Skill scored 96.8% diagnostic accuracy across 312 historical failure scenarios. Then number 313 showed up. The 450ms retry window was a compatibility shim I'd written five years earlier for RabbitMQ, and it got applied to a system that had been running Kafka for years. I'd even left a note in the migration docs: 450ms matches RabbitMQ GC window. Do not reuse outside this context. At the time, nobody understood it. They thought it was a stale config comment. Not until 4 AM, when the CTO went looking for it.
The AI didn't fail because it was wrong. It failed because it was right about yesterday — and yesterday wasn't running anymore.
Then I joined this company. And on the walls and along the corridors, there are two signs:
Your experience is waiting to be forged into an Agent. Great employees get the work done. Great Agents keep getting it done.
(In Chinese, since that's what they actually say: 你的经验，正在等待被炼成 Agent。／优秀员工完成工作，优秀 Agent 持续完成工作！)
Me: 15 years in QA, most of the titles in this field checked off at least once. By the standard of those two sentences, I'm not an employee. I'm raw material, still in the wrapper.
And look at that second one again. Great employees get the work done. Great Agents keep getting it done. So I do the doing, and it does the keeping. I raise it, it replaces me. Structurally speaking, the logic holds up.