2026.26: Nine Out of Nine Hundred
Happy Monday friends, this week from Las Vegas.
I sat in on an HBR virtual roundtable last week. Tom Stackpole hosting, with Kate Niederhoffer from BetterUp Labs and Tom Davenport from Babson. An hour on “the hidden realities of AI adoption.”
I went in expecting to disagree with something. I didn’t. Which is its own kind of uncomfortable, because they spent an hour saying the thing I’ve been writing in this letter for six months, with better data and a Harvard logo behind it.
The argument: AI adoption fails or succeeds less on the technology and more on the invisible stuff. Workflows. Trust. Talent. Whether leaders actually invest in augmentation or just say the word in a town hall.
Two things landed hard enough that I’m still chewing on them.
Let’s break it down.
Signal:
STOP TREATING IT LIKE A SOFTWARE ROLLOUT.
Davenport told a story about Johnson & Johnson that I haven’t been able to shake.
They catalogued 900 individual AI use cases across the company. Nine hundred. Every team, every Copilot licence, every clever little workflow somebody built on a Tuesday afternoon.
Then they threw most of it out and focused on 9 that actually mattered to the enterprise.
Nine out of nine hundred.
That’s not a cost-cutting story. That’s a measurement story. Davenport’s point was blunt: “the individual productivity, in general, I have found, doesn’t yield measurable productivity in most of the cases that I’ve seen — in part, because people don’t measure very well.”
You cannot measure 900 things. You can measure nine. And if you can’t measure it, you’re not running a strategy, you’re running a hobby with a budget line.
Here’s where it connects to something I wrote back in June. In 2026.21 — AI Gets You in the Game. Practices Keep You at the Table. I pulled a McKinsey number that I keep coming back to: redesigning the workflow is the single change most correlated with bottom-line impact from AI, and only 21% of companies have done it.
Twenty-one percent. The other 79% bolted a faster engine onto a cart with square wheels.
Now put that beside the MIT NANDA finding I’ve cited more than once here — 95% of enterprise AI pilots deliver no measurable P&L impact, after $30 to $40 billion invested.
Those two numbers are the same number wearing different clothes. The 95% failure rate isn’t a model problem. It’s the 79% who never redesigned the work.
And then Davenport did the thing that made me laugh out loud and then feel bad about it. He said when he gives talks, he asks the room to raise their hands if their company has a culture of disciplined measurement and experimentation.
In the United States, he said, you might get 1 out of 50.
One in fifty. That’s your real adoption rate. Not seats licensed. Not logins. Not the slide in the board deck.
Scale:
AUGMENTATION IS A LINE ITEM, NOT A VALUE.
The second takeaway is the one I think most leaders will nod at and then not fund.
Every leader I work with says the word “augmentation.” Nobody wants to be the person who says “we’re automating people out.” So augmentation becomes the safe word. It goes in the memo. It goes in the all-hands.
Niederhoffer named the gap precisely:
“I do think a lot of leaders are talking about augmentation. And the gap that exists is really in building the conditions for augmentation.”
Then she said the part that costs money.
A credible commitment to augmentation means investing in the talent infrastructure so heavily that it creates a relative decline in productivity at the beginning of the adoption curve. Because you’re paying for training. For psychological safety. For people to change how they work. For the time it takes to get good.
Read that again, because it’s the whole game. Real augmentation makes you slower before it makes you better. Which means it dies in every quarterly review where somebody asks “where’s the ROI.”
I’ve put numbers around this before. In 2026.18 — Your Team Already Priced In the Loss, I cited Deloitte’s 2026 Human Capital Trends: 59% of organizations are taking a tech-focused approach to AI, and those organizations are 1.6x more likely to miss their AI ROI targets than the human-centric ones. Microsoft’s Work Trend Index said the same thing in different math — organizational factors drive 67% of AI’s reported impact, individual mindset and behaviour drive 32%. Roughly 2:1 in favour of culture.
And in 2026.24 — You Took Your Team Hostage, the WalkMe survey of 3,750 workers across 14 countries: roughly 80% are avoiding or rejecting the AI their own company handed them. Only 9% of workers trust AI with a complex decision. Among executives, 61%.
That trust gap is not an adoption problem. It’s a receipt. The people who bought the tool believe in it. The people who have to use it every day do not, and nobody built the conditions that would change their mind.
Here’s the one that should worry you most, and it’s the one nobody has a plan for.
Niederhoffer flagged a reverse bias in her research: younger workers today trust AI less than senior workers do. Not more. Less. Every assumption you have about digital natives leading the charge is backwards.
And Davenport told a story from 2013 — thirteen years ago — where companies were already saying they didn’t need entry-level workers because AI could make those decisions. He asked them how they planned to get experienced workers in ten years if they stopped hiring at the entry level.
Their answer: “yeah, we haven’t really figured that out yet.”
They still haven’t. That’s the part that gets me. Thirteen years, and the answer is unchanged.
Judgment is not a training module. It’s reps. If you automate away every entry-level task, you have automated away the place where judgment gets built. And judgment is the only thing you’ll have left to sell when the tools are commodity — which they already are.
Deep Dive:
COUNT YOUR NINE HUNDRED.
No framework this week. One exercise, and it’s uncomfortable on purpose.
Step one. Get an actual count. How many distinct AI use cases exist inside your organization right now? Not the ones IT knows about — the real number. Every prompt library, every side-door subscription, every person quietly doing their job differently since March. If your answer is “I don’t know,” that is the answer, and it’s the same answer J&J had before they went looking.
Step two. Of those, how many have a baseline number you wrote down before the tool touched the work? That’s your real portfolio. Everything else is anecdote.
Step three. Pick your nine. Or your three. Or your one. Then say out loud what you’re not doing, so the 891 other things stop quietly consuming attention you never budgeted.
Step four — the hard one. Look at the nine and ask: for each of these, did we redesign the work, or did we just add a tool to the existing work? Be honest. Adding a tool is substitution. It’s the thing I called out back in March in 2026.10 — You Lost the Plot. Substitution is where digital transformation goes to die.
Then go find out where the ROI conversation is going to kill you. Because if you commit to augmentation properly, productivity dips first. If your board doesn’t know that going in, they will read the dip as failure and pull the plug at exactly the moment it was starting to work.
Tell them now. Not in Q4.
Here’s what stayed with me from that hour.
Nobody on that call said the technology was the hard part. Two of the most cited people in this field, an hour of airtime, and the technology barely came up.
It was workflows. Trust. Whether the junior person still gets to learn anything. Whether you measure the thing or just deploy it.
None of that is on a vendor roadmap. All of it is on yours.
If we get the human system right, the technology becomes an accelerant. If we ignore it, AI just helps us fail faster.
What’s your number — how many AI use cases are actually running inside your organization right now? Hit reply and tell me. Especially if the honest answer is “I have no idea.”
The newsletter isn’t the conversation. The conversation is the conversation.
See you next week.
Best,
JT
PS - If someone forwarded this to you and you want it in your inbox directly, subscribe HERE.
