ISSUE № 15 · September 21, 2026 · 3 min READ
The two ways AI rollouts actually die
The real failure modes have nothing to do with AI at all.
I'm Jeremy Hurst, VP of Autonomous GTM at Swan. I left a $450K VP Sales role to take a job with a made-up title. I did it because I got a glimpse of the 100x seller - it stopped being a thought experiment and became the actual playbook. This is a field journal for GTM and growth leaders who can feel the model changing under their feet and prefer changing with it to being left behind.
"Let's say we wake up 6 months from now and this project's been a complete failure. Why?"
A prospect asked me that yesterday during lunch. He'd already stated an intention to move forward, but wanted to pre-mortem the failure modes together.
I love the question - it's not one I get very often.
I gave him 2 reasons:
1 - Self-serve deployment. Swan in 2025 was pure PLG. No implementation or training. "Just talk to the onboarding agent," they said.
Somehow, many customers DID figure things out. But many others didn't. And when you ask why, it often boiled down to a lack of support.
No guidance on how to set things up, which use cases to tackle first, or how to integrate Swan with other tools in the stack.
We killed the gap only recently. You have 5+ users? Our forward-deployed motion's not optional.
2 - Doing too much at once. Some teams try attacking 5 or 6 use cases all at the same time.
Unsurprisingly, none move the needle. No measurable ROI on any of it, so the agents die quietly. Usage drops, and everyone defaults back to doing things the old way.
That's why we cap it - we attack 1 or 2 GTM motions max at the start of an engagement. Nail those first, then consider what comes next.
As I think about both reasons, neither has much to do with AI.
Support & change management. A disciplined focus on what matters at the expense of other things. These are very human challenges.
That's the part of AI transformation that actually decides outcomes. It's not the model or the agents. It's the people adopting them.
It's why we always talk about augmentation & never replacement - it's taking the judgement & expertise of humans and encoding it into the system, so you get 100x the output from your best people.
THE SIGNAL
Every rollout I've watched die, died for one of two human reasons, and neither of them was the model.
THE PLAY
The handoff
- Count how many people are actively using the agent today.
- If it's 5 or more, ask who owns onboarding for the next use case.
- If the answer is no one, build that support before you add another user.
SIT WITH THIS
The agents you've already deployed are either starving for support or buried under too much scope. Either way, that's not something a better model fixes.
COMMUNITY NOTES
The humans adopting it decide whether any of this actually works.
-Jeremy
Keep going