Who owns what on our GTM team now

Hey there,

Matteo here from Kinetyca.

I wanted to share something that has been on my mind lately. We are halfway through 2026, and almost every GTM team I talk to is running agents somewhere in their flow now. Reply handling, enrichment, reporting, something. It happened fast, and it happened quietly.

Here is the part nobody has actually done. Nobody has written down where the line sits. What the humans still own, and what the agents run now. The work moved off people's plates months ago, and the job description never got updated to match.

So I wrote ours down. Here is how the split actually sits on our team right now.

01What the GTM engineer still owns

The judgment work. The parts where being wrong is expensive, and a person has to answer for it.

• ICP and offer diagnosis. What is actually true about the market, not what the dashboard implies.
• System design. What the agent does, where it stops, and what happens the moment it is wrong.
• The messaging angle. The point of view behind the copy, not just the words on the page.
• Quality control. A person writes what the agent runs on, and nothing goes live until it has been run for real, not just reviewed on paper.
• Every live conversation. Objections, negotiation, the read on a room an agent cannot fake.
• The kill or scale call. The system computes the verdict. A person still pulls the trigger.

02What the agents run now

The repeatable work. The parts that used to eat a full day and now happen before anyone opens a laptop.

• The enrichment waterfall. Clay scores and layers signal before a lead ever hits a sequence.
• Reply triage. Claude sorts every reply against the SLA, suppresses the opt-outs, and drafts the next message for a human to approve, all before anyone opens the inbox.
• First-draft copy across every touch in the sequence, shaped by the style locked for that campaign.
• The kill and scale math. Reply-rate floor, sample size, deliverability guardrails, run against the real numbers every week.
• Deliverability health. The worst sender accounts get flagged and a fix gets proposed before a domain quietly dies.
• The weekly numbers. Week over week, month over month, and program to date, staged before anyone opens a dashboard.

03Why the line matters

Get this split wrong and it shows up fast, in both directions.

Push judgment onto the agent, and prospects feel it in the first reply. Keep a person running what an agent should already own, and you are paying a premium salary for data entry.

Here is the part the average AI agent pitch leaves out. A lot of what looks fully automatic still has a person approving it before anything real goes out. That approval gate is the whole point. It is what keeps the speed of an agent without losing the judgment of a person.

This line moves every few months as the agents get better. Right now, this is where it sits on our team.

I would genuinely like to know where it sits on yours. What has moved to an agent, and what you would never hand over.

Hit reply and tell me, I read every one.

PS. If you want to map where this line should sit in your own GTM, grab 30 minutes with us here. That is the fastest way to get another set of eyes on your setup.