AI adoption statistics in public are a costume. On stage, founders talk about transformation, copilots, and the percentage of the roadmap that is now “AI-first.” Off stage—after the cameras, after the branded water, after the off-the-record nod—the language changes. It becomes operational. It becomes about review queues, customers who will not let data leave a region, and the difference between a demo that closed a round and a workflow that closed a ticket.
This Field Notes piece is about that register change. I will not invent a panel of composite characters with fake names and fake quotes. I will describe patterns I have heard repeatedly from founders, product managers, and technical leads who asked not to be the anecdote. AI industry commentary that only uses the on-stage register is fan fiction with better lighting.
The On-Stage Script Versus the Working Script
On stage, adoption is a headline. Off stage, adoption is a definitional fight. Does a support tool that drafts replies count if agents still rewrite every one? Does an internal hackathon count? Does a customer who enabled a feature and then turned it off after a compliance review count as churn or as wisdom?
When the cameras are off, people get more honest about measurement because they have to run a company, not a keynote. They talk about shadow use: employees pasting sensitive text into consumer tools. They talk about the opposite: employees refusing a mandated copilot because it slows them down. Both can exist in the same firm in the same week.
Themes that recur when the recorder is down
“We counted seats because we could not count completed work.”
“The customer bought the story and then asked for an air-gapped version we do not have.”
“Our own engineers use the tools more than the department we sold them to.”
“The model is fine. The permissions model is the product.”
“We stopped saying autonomous after the first time it emailed the wrong list.”
Those lines are paraphrases of a type, not a gotcha aimed at a named person. The type is the story.
Why Off-the-Record Honesty Is Rational
Founders are not villains for having two registers. Investors, reporters, and recruits reward the first one. Customers and employees punish the second if it is too far from reality. The gap is a survival tactic until it becomes a product-quality problem.
I have been on both sides of the notebook. At WIRED and Fast Company, I needed a sentence that could survive an editor. Independently, I can print that the sentence was the least interesting thing said after the mics went cold. Read the announcement. Then read the incentives. The on-stage incentive is narrative coherence. The off-stage incentive is not to lie to the people who have to implement the slide.
What I treat as more informative than a launch quote
Public artifact | Off-stage counterpart |
|---|---|
“AI-powered platform” | Which job is actually on the critical path |
“50% of employees using” | Weekly completed tasks, not logins |
“Enterprise-ready” | SSO, logs, DPA, and a named security contact |
“Agent” | List of tools it may call and who can revoke them |
“Saves hours” | Hours added in review and incident response |
The right column is where AI adoption statistics should live. They rarely do, because the right column is slower to gather and harder to brand.
How I Conduct These Conversations
I still carry a paper notebook. People talk differently when they are not watching themselves on a laptop screen. I offer to keep specific product details off the record and to publish patterns. Some founders still perform. Some seem relieved to drop the performance. The relieved ones tend to have a clearer product.
I ride the subway or a bicycle home and write the last-page question: who really benefits, and who really pays? Off stage, founders will sometimes answer it. The people who benefit from the hype cycle are not always the people who will have to staff the exception queue. The people who pay are often junior employees asked to “just try it” on top of their existing work.
Rules I use so this does not become gossip
No fake names, no reconstructed dialogue presented as tape.
No single anonymous source carrying a scandalous factual claim.
Patterns need independent public corroboration when they imply a market fact.
If a founder wants to go on the record later, they can; the pattern still has to stand without them.
The off-stage register is for operational truth, not for score-settling.
Those rules make for less spicy copy. Spicy copy is how you get sued and how you stop getting the second conversation.
What Readers Should Hear in the Quiet Version
Here is what changed, and what did not. More founders now have something real to sell, which makes the off-stage conversation more technical and less purely visionary. The dual register remains. If you only consume the camera version, you will think adoption is a switch. If you listen after the lights dim, you will hear a stack of partial uses, compliance delays, and a few workflows that actually moved.
That mix is the adult version of generative AI business news. It is also why this blog exists. I will keep going to the talks. I will keep staying for the hallway. The hallway is not the secret truth of a cynical universe. It is simply closer to the work.
No notes on this sheet yet.