The GTM engineer is a systems thinker with a job title

A GTM engineer is someone who builds the system that does the marketing, instead of doing the marketing. That is the whole job. Clay, the company that named the role in 2023, defines it in one line I have not been able to beat: the unit of work is a system, not a task.

In August I wrote that I was rebuilding OpenHuman’s marketing as workflows, not campaigns. A campaign ends. A workflow keeps running when you’re asleep. I thought I was describing a personal preference. It turns out I was describing a job title, and the industry got there before I checked.

So this essay is me doing the honest thing: holding my own idea up against what the GTM engineering people have actually figured out, and seeing where each side is further ahead.

1. A task is a flow. A system is a stock.

Systems thinking has two basic objects. A stock is something that accumulates, like water in a tank. A flow is something that moves, like water through a tap.

Most marketing work is flows. You send the emails, you post the launch thread, you run the webinar. The moment you stop, the motion stops. Nothing accumulated except, hopefully, some pipeline.

A workflow is different in kind, not in speed. Clay’s guide puts it precisely: the engineered version “is not faster manual work. It is a different category: the rep was doing the research, the GTM engineer builds the thing that does the research.” The rep’s output disappears when the rep goes home. The workflow is a stock. It sits there, and it keeps producing.

Think of two villages by a river. One hires more and more people to carry buckets. The other spends a hard month laying a pipe. For that month, the bucket village wins on every metric. Then the pipe opens.

2. Why the headcount math broke

The old revenue model was linear: more reps, more emails, more pipeline. You scale output by scaling people. For about fifteen years, it worked.

Two things broke it. The tactics got commoditized, so a templated cold email now performs like what it is. And AI collapsed the cost of building the alternative, so one person with Clay or Claude Code can assemble a research-and-outreach system that used to need a team. Mistral AI mapped 25,000 qualified enterprise accounts in two weeks by chaining ten data providers into one workflow.

Clay’s guide compresses this into a sentence that should be on a wall somewhere: “Your GTM motion isn’t under-staffed. It’s under-engineered.”

The systems translation: hires are a flow you pay for every month. Workflows are a stock you pay for once and maintain. One compounds. The other doesn’t.

3. The loop that is eating outbound

Here is where it gets properly systems-y, because automation has a dark loop in it, and the loudest warning came from inside the sales world.

Jason Lemkin wrote in 2024 that AI would be the death of outbound. Not because the AI is bad. Because of what managers do with it: “set and forget” cadences at 100x volume. His line was blunt: “Slightly better than pre-AI for sure. But 100x more of them will on balance make them even worse.”

Draw it as a loop. Cheaper messages lead to more volume. More volume leads to lower reply rates. Lower reply rates lead to even more volume to hit the same number. That is a reinforcing loop, and it runs until it slams into the balancing loop nobody tracks on a dashboard: buyer attention. Attention is a stock too, and the whole industry is draining the same tank.

This is the test that separates a GTM engineer from someone who just automated the spam. The engineer sees both loops. The volume operator sees only their own flow.

4. The honest part: the title is ahead of the hiring

I am not going to tell you this is the hottest job of the decade, because the data says it isn’t yet.

Kyle Poyar went looking for the hiring wave in 2025 and found roughly 45 GTM engineer job posts in a month, one for every 92 SDR posts. His conclusion: “I bet there were more LinkedIn posts last month about GTM engineering than actual companies hiring for it.” Nearly half the people with the title are agencies or consultants. Brendan Short, who has championed the role since 2024, says it himself: “The conversation is (currently) outpacing the adoption.”

I don’t think that kills the thesis. New system-level roles always get named before they get headcount. DevOps looked like a LinkedIn fad once too. But it does mean the honest claim is smaller: the title may or may not survive, the work will. Someone at your company is going to be the person who builds the thing that does the work. The only question is whether they do it on purpose.

5. How you can do it

You do not need the title to do the job. Here is the version I am running at OpenHuman, corrected by what the GTM engineering playbooks got right before me.

Give every recurring job an owner, an input, and an output. If you cannot name all three, you have a vibe, not a workflow.

Fix the data before the cleverness. Clay’s three rungs are foundation, modeling, activation, in that order. Most teams fail by running smart plays on dirty data. I’d been doing activation-first. They are right, I was wrong.

Trigger on state changes, not on calendars. The best teams fire a play when something happens in the data: a signup, a champion changing jobs, a usage spike. A calendar says “it is Tuesday.” A state change says “something is actually different.”

Switch workflows on one at a time. You are introducing loops into a live system. Introduce them slowly enough to see what each one does.

Keep a human where the stakes are. The operators I trust automate the small-deal motion fully and keep people on the big-deal motion. Know which tank you are allowed to drain.

In short

• GTM engineering is systems thinking applied to revenue: build the thing that does the work.

• Hires are flows, workflows are stocks. Stocks compound.

• Automation contains a reinforcing loop that destroys the channel it runs on. Watch the balancing loop: buyer attention.

• The hype is ahead of the hiring, and that’s fine. The work is real even while the title is unproven.

• Owner, input, output. Clean data first. Trigger on state changes. One workflow at a time.

The discipline is the same one I keep arriving at from every direction: take care of the system, and the output takes care of itself. Identify the one recurring job you are still doing by hand, give it an owner, an input, and an output, and switch it on. Then do the next one.