Five mornings, five versions of the same underlying problem. The accountability an organization runs on almost always lives one floor below where the executives sit, and the company pays whenever the top mistakes its own vantage point for the whole view.
Microsoft spent the first days of July resetting its Xbox business. Asha Sharma, who took the division over from Phil Spencer in February when he left the company after thirty-eight years, told staff plainly that the business “is not healthy” and warned that Xbox had “spread ourselves too thin.” The reset came with about thirty-two hundred jobs cut, four studios leaving, and Helen Chiang promoted to a new chief operating officer role with end-to-end profit and loss across content, hardware, platform, and services. Set that memo beside the record it is answering. Xbox had reportedly spent more than twenty billion dollars over five years while the revenue underneath drifted down, and nobody with the authority to change course seems to have sat close enough to the work to feel the gap before it reached this size.
That is the pattern my five pieces this week kept circling from different angles, whether the subject was a leadership team modeling the very silos it told everyone else to break, an AI tool nobody was answerable for, an executive who had drifted from the floor where the work happened, a strategy chief handed responsibility without the authority to decide, or the plain finding that leaders rate their own accountability well above what the people below them see. A reset is the moment a company admits, publicly and at real cost, that the view from the top had gone stale.
Sharma also capped the studios at five management layers, which is the one move in the announcement that actually pushes a decision nearer the work. The thing I would watch is whether the rest of the new structure follows it down, or whether it mostly redraws the top of the chart with the word accountability written across it.
A fifty dollar lesson in who an AI feature is built to serve, and what the people paying for it are owed.
I pay LinkedIn about fifty dollars a month, and for years I could not have told you exactly what for. So when its new AI job search showed up, I opened it half hoping this was the thing that finally earned the bill. There was nowhere to type. What it offered was a row of pre-written buttons, eight or so, each a generic wish that plainly belonged to someone in some other line of work than mine. I clicked the nearest one and it went looking. Welcome to the future of job search, the screen said.
What struck me was what it would not let me do. LinkedIn holds more than fifteen years of my working life, the profile it prompts me weekly to make more complete. The obvious build, the thing the technology is finally good enough for, is to read that history and tell me what I am suited for, including the roles I would never think to search and the lateral move I would dismiss on the title alone. What I got instead was a short menu of canned options that fit no one in particular, and me least of all.
Afterward I found LinkedIn’s own help page for the feature, and it describes something better than what I saw, a field where you type the job you want in your own words. Maybe that version was still on its way to me, since the page says not everyone has access yet. Either way, the same page draws one line in plain language. The search, it says, does not support looking for “jobs I’m qualified for.” Read that twice. They built an AI to help you find work, then noted in the manual that the one thing you would most want from it is the thing it will not do.
When I went back to find even the button version for this piece, it was gone from my menu. The feature has not been retired. LinkedIn says more than a million people use it every day and it is spreading across languages and markets. It simply is not showing for me right now, and nobody told me why, or when it left, or whether it returns. I pay the same fifty dollars either way.
Hold those two things together, because the gap between them is the argument. What I think I am buying is a tool that helps me, the subscriber, solve my problem. What the company built the AI to do is a different thing, and you can see it once you look at who actually pays LinkedIn. My subscription is not the business. Talent Solutions, the recruiting product, is LinkedIn’s single largest revenue line by a wide margin, well ahead of the subscriptions like mine. A single Recruiter seat runs well into five figures a year, and what it buys is the right to search more than a billion profiles and reach people who are not even looking. I would bet the AI budget followed that same money, into the automated recruiter tools that now source and screen candidates on their own. The person on the other side of the search got a tool that asks what he wants and keeps his profile fresh for the recruiters paying to find it.
I do not think LinkedIn built a bad feature. I think it built the honest one for its real customer and handed the rest of us a friendlier front door on the same machine. That is a defensible business. Charging an individual fifty dollars a month and telling him almost nothing true about what he is getting is harder to defend. If you sell a person a tool, you owe them a plain account of what it does and whether it works. You owe them a word when it appears and vanishes from their screen. Welcome to the future is marketing copy. A tool you cannot find and cannot measure is one you are paying for on faith.
And this is worth saying plainly, because it is not really a LinkedIn story. LinkedIn is the clean case, the one where the user and the paying customer are two different people on opposite sides of the same platform, so the mismatch is easy to see. Across a lot of products this year, AI showed up to make a company look like it was keeping pace, for its board and its investors, more than to solve the problem the user actually had. Most of them hide it better than a job search that offers you eight buttons and none of them fit.
I spent years as a product owner, and this is the oldest tension in the work. The feature that would most help the user and the feature that serves whoever signs the big check are usually two different features, and the second one wins, whatever the roadmap slide says about being user obsessed. AI did not change that tension. It gave everyone a faster, shinier way to ship the second feature and call it the first.
So if you are putting AI in front of people who pay you, two questions are worth answering before the launch post goes out. Did you aim it at their hard problem, or at the one that is easiest for you to keep serving? And if they hand you money every month, have you told them the plain truth about what the thing does and whether it works? People can feel the difference even when they cannot name it. I feel it every time I open that app and wonder, still, what I am paying for.
Next up on High Road Conversations is Hanna Bauer, who survived a childhood of heart failure and two heart attacks and learned to become friends with uncertainty, then carried that through years running a publishing company while print collapsed under her. Her episode arrives Monday, July 20. Subscribe on YouTube so it reaches you the day it goes live:
Listen →
I don't just talk about Agile and leadership, I put it into practice in my Skool, building and shipping real products with Claude every Thursday at 12:30pm EST.
Join us →A chapter in Agile Sucks! (When You Do It Wrong) follows five civilian contractors in a combat zone who took a monthly flood of hostile incidents, more than eight hundred, down to under fifty in a year by pushing accountability out to the people closest to the problem instead of holding it at the top. It is the clearest case in the book of what this week kept circling, that ownership only works when it sits where the work does.
Get the book →A short read every Saturday morning. No spam, unsubscribe anytime.
Subscribe