In software, there is a word for the thing that breaks your system. We call it an edge case. An edge case sits at the very far end of a system's operating range. Think of the customer's last name that's too long for the field, an order that arrives 1 second before midnight on the last day of a leap year. Now imagine everybody that a system is built to serve lined up shoulder to shoulder. Most everybody is bunched in the middle. That's who gets designed for because that's
where the numbers are. Now go to the end of that line. That is the edge case. 30 years ago, I stepped inside my first server room. The floor was a lid on top of a nest of cables. There was a wall of noise that you had to talk over. And in the corner of the room, there was a terminal with a green screen flashing like it had all the time in the world. And as I stood in that doorway, I did the arithmetic I would do
in every room like it for the next 30 years. I counted. I was the only black woman it. I didn' just build for edge cases. I was one. For a long time, I thought that was purely the cost. And to be honest with you, a lot of it was. Being an anomaly is very expensive. It's a thousand small negotiations that you run every day that nobody else in the room is running. But it also handed me an instrument. One that I wouldn't understand for years. You
see, when you sit in the middle of a system, it's like the weather. It's the conditions. You don't notice it more than you notice air. But when you sit at the edge, you notice its shape. And you see where it's designed to stop. So I stopped reading the dashboards. Dashboards are written for the center by the center. They tell you that there is 92% adoption. But what they don't tell you is about the woman who comes in 20 minutes early every single morning because the badge
reader doesn't recognize her. So I started reading what I think of as the exhaust, the workarounds, the shadow spreadsheet someone creates because the real tool doesn't work. Now, to be honest, that's not a complaint. That spreadsheet is a blueprint. The meeting after the meeting. The exhaust is where a system tells the truth about itself. And here is the truth that I have learned in 30 years of sitting at the edge. And the one thing I'd like for you to leave here with today. The people a
system fails are not an exception to the system. They're the most accurate instrument you have for reading it. And it would take a very expensive failure for me to learn to listen to that system. Eight years ago, I led a rollout. A large one. Was, you know, millions of dollars, thousands of users. And we rolled it out on time. We closed the project. Somebody bought a cake. It was technically clean. And we were all good when it went out. But there was one person during the
test run, the small thing you do before you build the real thing. She wasn't confused. She knew it faster than most people in the room. Her question was, what would it cost me? What would happen to the parts of my job that no one would be able to see? She asked me three times on three separate occasions. And I logged her as a resistance to change. There's a really funny thing about that phrase. I've thought about it for years, resistance to change. It's one of the
most dangerous phrases in corporate life, because it allows you to take a signal and file it as a personality problem. Six months later, this platform I rolled out was a ghost town. Nobody revolted, nobody complained. They just quietly went back to the spreadsheet, because the spreadsheet could still hold all of the parts of their job that I had determined didn't exist. And here's the thing I would rather not tell you. For 30 years, I was the one a room decided was a complication. I was the
one whose words were heard as friction. I could tell you exactly what it felt like from the inside. I could describe it to you in my sleep. And the first time I got the budget, the authority, and the pen, I did it to somebody else. I hadn't built a bad tool. I had built a tool for the center of the curve, and I'd labeled the edge of the curve a personality problem. I had the instrument. She was in the room, in the test group, on the
record three times. And I filed her. My system didn't fail because of an outlier. My system failed because I noted her as an annoyance. But she was the best source of truth in the room. Now, I could tell you that culture is the reason for all failures. Sometimes the tool is genuinely bad. And sometimes the strategy wasn't good before anyone wrote a line of code. But in 30 years, I have never seen one of these things fail because of the code. They died because of everything
around it. And this is why it matters this year more than it has a decade ago. Technology has never given us our worst habits. What it does is scale them at whatever speed it runs. A bad manager with a spreadsheet damages a team. A bad manager with an automated system damages a team on schedule permanently before anyone has time to notice. And the people at the edge feel it first. They always do. In the American Psychological Association's 2023 Work in America survey, four out of 10
workers said they were worried the AI would take some, if not all, of their job responsibilities. 64% of those worried workers said they were tense or stressed the majority of the workday compared to 38% that weren't. We have been using humans as a gap, a place to patch the problem for the past decade. This is what happens in workplaces often. There's a person on every team who absorbs the confusion, translates the strategy, works hard, holds it all together. We have been using people as human shock
absorbers. Shock absorbers have a speed they're rated for. Beyond that, they break. This year, I had another chance. Not a rollout this time, a rule. I decided my whole team was going to learn AI properly. 3 months of classroom training on the calendar, mandatory. I had the budget, the authority, and the pen. And this time, I was going to use it for good. And my team told me it would never work. I want you to know how close I came to hearing that is resistance. It
has the same shape. Somebody slowing down the thing you're certain is going to be good for them. But this time, I asked the question I was too busy to ask before, what will it cost you? And they told me that it was going to be something they didn't have. It was 3 months out of the work while the work kept arriving. And then they actually told me what they needed. It wasn't a class about AI. It was training inside their job, role and responsibility that they
did on a regular basis. That was the reading. I didn't have to be clever. I just had to hold still long enough to take it So I stopped designing the training and I let them design it. And what they built looked absolutely nothing like I would have put on the calendar. It was small, it was consistent, and it lived inside their day-to-day tasks and skills. And once it was up and running, they told me it saved them 10 hours a week in their words, not a
spreadsheet. Hours they could now contribute to higher priority work that got pushed out before. That's the whole method. There isn't a framework. There isn't an acronym. Find the people your plan would cost the most. Ask them how much it will cost. Believe the answer. That is what the instrument is for. Now I used to think this conversation was for boardrooms or for people like me who have budgets and rollouts. It's not anymore. An algorithm determines what shifts a 19 year old gets and whether those hours
qualify for benefits. An online tool has a checkbox for family and half the families in the district don't fit it. And this morning a piece of reading software evaluated a classroom full of children to see if they were on track. There's a girl in that class. She's dyslexic. She's not behind. She's the most accurate reading that piece of software will ever have. Now no one in any of stories is a villain. They just created tools like perfectly reasonable people for the center on a deadline. That's
all it takes is a deadline. So whether you're running a company, a classroom, a clinic, or a kitchen table, you are designing systems whether you think you are or not. And you know who they fail. There was someone who came to mind while I was talking. There always is. I don't want you to go fix that person. I don't even want you to just include them. I want you to read them. Go ask the question I was too busy to ask. What will it cost you?
And believe the answer. I never got to ask that engineer from eight years ago, the one who asked me three times, because when you build for the middle, the edges break. But when you build for the edge, the middle comes free. 30 years ago, I stood in a doorway and I counted. And I felt counting was the price for being in the room. It was never the price. It was the reading. And that person who came to mind for you is not your exception. They are
the people your system fails the most and the most accurate instrument you have for reading it. >> Thanks. >> [cheering] >> This girl is on fire.