September 30 was a strange day to be me. Two of my worlds proved the same thing on the same day — and both of them proved it about the thing I do for a living.
First, the world I work in by day. ZoomInfo's CEO, announcing an acquisition, said the quiet part out loud:
"Every revenue leader has seen an AI agent demo. Almost none have seen it actually run a real GTM motion end-to-end."
Gong, same day, at its own conference: "Enterprises are no longer satisfied with evaluating AI; they want scalable results." And HubSpot had already renamed its entire strategy around the "outcomes era" — customers want "proof of value in their own environments, not feature lists."
Second, the world I work in by night. Instructure — the company that owns Canvas — published its first-ever analysis of what 19.5 million US college students actually plugged into their LMS. The top tools are the boring ones: McGraw Hill textbooks, Zoom, Kaltura. The most-adopted dedicated AI tool, Google Gemini, reached 87,788 users across 211 institutions and didn't crack the top 100. The textbook plugin hit 11.4 million. Institutions signed the enterprise deals. Nobody embedded the tool in the class. Only 11% of instructors got comprehensive AI training.
Two industries, same day, same sentence in two vocabularies:
Buying AI is a checkbox. Deploying it is the moat.
Here's the part I'm still not sure is a flex or a confession. I'm the rare person whose two jobs sit on the two sides of that exact gap.
By day, I run sales ops at emergence.ai. I'm the buyer those vendors are now courting — the one being asked for "proof of value in my own environment." My literal deliverable isn't "which agent should we buy." It's "which number moved, by Friday, and did it survive."
By night, I'm a professor of practice teaching GenAI to business students. I'm inside that 11%. I don't hand my students a license and hope. The rule in my classroom is "build, don't describe" — every sprint, they ship something real. And I already wrote the discipline down before the market had a word for it: the delivered session is ground truth, the plan was a hypothesis.
So I don't get to evaluate AI. I have to make it run. In both jobs. That's not a brag. It's just what the rent is made of.
And here's the part that made me laugh. The vendors spent September 30 admitting the demo is the easy part — and then each of them announced a product to sell the fix.
Gong's answer to "enterprises want results, not evaluation" is a product called "Gong Activate." Apollo's is "the AI GTM system." HubSpot's is "outcome-based pricing" — literally charging for the outcome it can't guarantee. ZoomInfo's is buying a company to make agents "reliable enough to run GTM motions."
They can sell you the loop. They can sell you the system. They can even sell you the pricing model. What none of them can sell you is the Friday afternoon — a rep actually changing their workflow, a professor building the tool into a course. That's deployment, and it has no launch event.
So I've stopped asking "is this demo impressive?" and started asking four questions instead. They're the only four that separate "we bought the agent" from "a number moved":
- What changed in a human's workflow? Not "what did the agent do" — what does a rep, professor, or student now do differently?
- What's the counterfactual? Against what baseline? "Pipeline went up" isn't measurement.
- What number moved, and did it survive? The one that's still true next quarter, not the demo-week spike.
- What would make us turn it off? If you can't name the off-switch, you deployed a pet, not a tool.
I didn't crack AI deployment. I'm not ahead of the market. I just never had the luxury of stopping at the demo — my day job is a number by Friday, and my classroom is a cohort that has to ship. The market spent one day this week discovering, from both directions at once, that "make it run" is the scarce skill. I've been forced to practice it twice over, because I had no other option.
The demo was never the hard part. The deployment always was.
And the question I keep turning over isn't "which agent do I buy." It's the same one the vendors are suddenly desperate to answer for me: if deployment is the moat — and nobody can sell it — then who in your org is actually on the hook to make it run?