Pick a launch motion — then move faster than feels comfortable
Every AI product launch is a bet on how you'll learn. The three motions I reach for each buy a different kind of signal, at a different cost:
| Motion | What it buys you | What it costs you |
|---|---|---|
| Closed pilot | Deep, high-trust signal from a handful of design partners; safe to be rough. | Slow; tiny N; you risk over-fitting to a few loud voices. |
| Open beta | Volume signal and market momentum; surfaces use-cases you never imagined. | Support load; reputational risk if the experience is half-baked. |
| A/B testing | Causal proof of what drives value — and what people will actually pay for. | Needs real traffic and clean instrumentation before it's trustworthy. |
I've run the full arc in practice — the Campaign Creation Agent went from a 3-customer closed pilot, to a 100+ customer beta, to 500+ GA accounts. Each stage was deliberately a different question: Does it work? Does it generalize? Will they pay?
Why the urgency is real
The reason you must move faster than ever isn't hype — it's that the model itself is no longer the moat. When every competitor can buy the same frontier models and the same tooling, the differentiator collapses onto how fast you learn and how well you apply your own context. Harvard Business Review put it cleanly:
"When everyone has access to the same AI models, the same AI-enabled tools, and the same vendor ecosystem, organizational context becomes the differentiator." — Rohan Narayana Murty & Ravi Kumar S, Harvard Business Review, Feb 2026
Early pilots and tests aren't just product validation — they're how you find your willingness-to-pay signal and the specific slice of value customers find compelling enough to open the wallet for.
The race to the bottom: vibe-vs-buy
Here's the uncomfortable part. AI and agents are a race to the bottom, and monetizing raw usage is not enough. The old "build vs. buy" calculus has become "vibe vs. buy" — it is now so easy to generate working software that an SI, a third party, or even a motivated power user can stand up a passable version of your feature in a weekend.
35%
of teams have already replaced at least one SaaS tool with a custom build
78%
expect to build more of their own tools in 2026
Source: Retool, "The build vs. buy shift," 2026
If someone can vibe-code it, your usage meter won't hold. The only durable position is an outcome and value set so compelling — backed by governance, scale, and trust — that it's not worth anyone's time to rebuild it themselves.
My pricing stance: monetize at the appropriate scale
Buyers — especially the C-suite — do not like committing to an ecosystem where the quantities are unknown. I hear a version of this in nearly every deal:
"I don't want to put agent limits on my lines of business — but how do I avoid overages, or runaway scaling, or getting surprised as we discover new use-cases?"
That's a fair concern, and it's usually met with unfair positioning. Shifting hard-right to full outcome-based pricing sounds elegant, but it's genuinely difficult: outcomes are caveated and different for every single customer. The answer lives in the middle — monetize at the appropriate scale.
- Give LOB leaders runway. Let them prototype and build rapidly, without limits, so they can actually discover where the value is.
- Pressure-test the patterns. Watch where in the platform the new agent behaviors actually unlock value at scale.
- Then price where it counts. Meter the things the agent uses most often, and the agent itself — not a punitive cap on the customer's ambition.
It's the same discipline I applied to the real packaging decisions on Agentforce for Marketing Cloud — onboarding GPT/Prompt licenses with Office-of-the-COO approval, building Finance-approved cost-to-serve models, and leading the migration off legacy Einstein Request entitlements to Flex Credits. Packaging is where strategy meets the invoice; get it wrong and the best product in the category still stalls.