Where Are Your AI Superpowers Actually Hiding?

Don Shin’s warning about vendor lock-in has a flip side: where are your AI superpowers actually hiding?

He identified six places where AI agents create disproportionate value:

Volume (repetitive tasks with variations) Assembly (prep time > actual work) Orchestration (routing between systems) Vigilance (continuous attention) Speed (quick responses matter) Access to expertise (bottlenecked by experts)

Most organizations try to solve vendor lock-in by replicating frontier models everywhere. That’s the wrong move.

Instead, find ONE or TWO places in your martech stack where one of those superpowers matters most. That’s where you implement.

Maybe it’s automating audience segmentation (volume). Or generating campaign variations based on performance data (assembly). Or routing leads between systems based on fit (orchestration). Pick the place where AI creates the most obvious value.

That’s also where open-source SLMs actually work. Because you’re building a narrowly-focused system for a specific superpower, not a general-purpose replacement for Claude. The model doesn’t need to be fluent at everything. It needs to be excellent at that one thing.

And the vendor lock-in payoff is that you own that infrastructure. Your segmentation logic, your routing rules, your data. No vendor can change your pricing or terms around the thing that actually moves the needle.

Own what matters, not everything.

So the question isn’t “should we replace Claude with open-source?” It’s “where do our AI superpowers create the most value, and do we want to own that?”

What’s your martech superpower? Where does quick AI intervention actually change outcomes?