How Real People Are Using Artificial Intelligence to Fix Things That Matter. Josh Tyrangiel.
Hey family. JP Morgan put this book on their list for students and interns. I am not going to turn this into a sermon. It is not a church book. But that list got my attention. If they want their young people reading it, I want us looking at it. Here is what I think they want you to walk away with. And here is what it hit in me.
Josh Tyrangiel writes for The Atlantic. He edited Bloomberg Businessweek and has won twelve Emmys and a Peabody as a producer. For this book he did not sit in labs with the people building the models. He went inside the institutions using them: the Cleveland Clinic, the IRS, an under-resourced Indiana school district, and Pentagon war rooms.
He was looking for what happens where the data is messy, the incentives are misaligned, and everything that matters is charged with human emotion. What he found is that the wins are quieter and more human than either the hype or the panic would tell you. Here are the key points.
AI should sharpen human judgment, not remove it. Every success story in the book keeps a human in the loop.
The professional has to be able to review the output and understand it. A black box nobody can question does not get adopted in a hospital or an agency.
Teachers, doctors, and public servants know where the work breaks. At the Cleveland Clinic the pilots are run by doctors, not the technology department.
A year of sepsis prediction software cut sepsis deaths 41 percent, roughly a thousand people, because doctors argued with the output.
Perna used a crisis to suspend the friction. Werfel worked quietly under the radar. Both moved. Neither waited for permission.
Tyrangiel says it plainly. He is pro-AI and anti-nonsense at the same time.
Stay passive and the people who are not thinking about the public good will design the technology for everybody else.
Tyrangiel walks into the Cleveland Clinic expecting to find the AI work happening in a technology department. It is not. The pilots are being run by doctors.
He finds the same pattern in an under-resourced Indiana school district, where a teacher, not a software company, figures out what AI should actually do in her classroom, and she sees it as protecting human connection rather than replacing it. His argument runs through the whole book: teachers, doctors, and public servants know where the work breaks. The technologists do not.
The best ideas in our proposals come from the people who actually run the network at two in the morning, not from anybody sitting in our office. So I start a technical approach by talking to the person doing the job, then I write. Federal evaluators can tell in one paragraph whether an approach was written by somebody who has been in the building.
The idea for a message is almost never mine. It is already sitting in the text, and my only job is to be close enough to it, often enough, to notice. Somebody who studies about the Bible will miss what somebody who lives in it every morning sees.
The Army taught me that the sergeant who has done the job for twelve years knows exactly where the plan will break, and the officer who does not ask him finds out the expensive way. I made a habit of asking the most experienced person in the room what I was missing. That one habit outperformed every school I attended.
My kids know things about their own world that I cannot see from where I sit. When I stopped diagnosing and started asking them what the actual problem was, the answers got better fast. Being the parent does not make me the expert on their day.
The Cleveland Clinic ran sepsis prediction software across its hospitals. Over a year, deaths from sepsis dropped 41 percent, roughly a thousand people who lived.
It did not happen because the software was turned on and left alone. Doctors got the most out of it when they compared its call against their own clinical experience and pushed back on it. Augmentation over replacement, with a human who can still review and understand the output.
I use AI on every proposal now, and I have never once let it decide. It drafts, it checks compliance, it sorts an RFQ pile I could not read by hand. Then I read every line myself, because the thing that gets a company disqualified is always the sentence nobody checked.
I let the tools do research, comparison, and packaging. I do not let them tell me what a passage means. The moment I outsource conviction, I have nothing left worth giving anybody on a Monday morning.
The Army beat the after action review into me, where you put what was supposed to happen next to what actually happened and refuse to look away from the gap. That is exactly what those doctors were doing with the algorithm. The value was never the output. The value was arguing with the output against real experience.
Isabella and I do not decide anything that matters off a single source, and a confident machine is still a single source. We look at it, we talk, we sleep on it. A tool can hand me an answer, and it cannot carry what that answer costs our house.
This is not a sales pitch for AI. Tyrangiel says flatly that it is not riskless and it is not a silver bullet for entrenched problems, and Kirkus called the book a sharply observed critique of AI hype.
He reports from places where the data is messy, the incentives are misaligned, and everything that matters is charged with human emotion. He is pro-AI and anti-nonsense at the same time. That is the harder position to hold, and it is the only honest one.
Half my industry is selling AI as a miracle and the other half is warning it will end us, and both crowds are avoiding the work. In federal IT I have to state exactly what a capability does, what it costs, where it fails, and who is accountable when it is wrong. Neither the hype man nor the doomer can write that paragraph.
Fear and hype are the same laziness pointed in different directions, and I will not teach from either one. If I am nervous about something I have not studied, then what I owe people is study, not a warning.
I have watched careers get made by the person who took an honest position on a new thing while everybody else picked a side and got loud. Be the one in the room who can separate what is proven, what is promising, and what is nonsense. That person gets called back.
When my kids bring me the extreme version of something they saw online, we go look it up together. Not panic, not a shrug. Teaching them to check is the entire skill.
Two government stories sit side by side. General Gustave Perna, running Operation Warp Speed, used the crisis to suspend the normal bureaucratic friction and rebuild how vaccines got distributed.
Danny Werfel, as IRS commissioner, did the opposite: he advanced his agency's AI capability quietly, under the radar, inside one of the most rule-bound institutions in the country. Crisis or quiet. Both of them moved. Neither of them waited to be handed a clear path.
Federal contracting is a rule-heavy world, and the winners are never the people complaining about the rules. They are the ones who read them closer than everybody else and find the legitimate lane nobody bothered to look for. Every certification and vehicle we hold is a door that was standing there the whole time.
I never got permission to start. I started teaching a message every weekday into whatever channel I had, on a schedule I set myself, and years later that discipline is the thing I am known for. Nobody hands out platforms.
The Army showed me there are two ways to move something heavy: a crisis that suspends the usual friction, or quiet consistent work while nobody is watching. Both are real and both work. Standing still until someone clears the path is not on the list.
Isabella and I have never once had ideal conditions. We built what we built inside the money, the schedule, and the season we actually had. Waiting for the right conditions is how families spend a decade planning nothing.
One of the researchers in the book studies vocal patterns in autistic children who cannot speak. One of those children is her own son. She did not wait for the world to decide her problem was worth funding.
Tyrangiel puts the same charge on the reader: actively demand the AI systems you want, because if you stay passive, the people who are not thinking about the public good will design them for you.
I stopped waiting for a vendor to build what my company actually needs. We describe the workflow we want, then we go make it. Hand that design job to somebody else and you get their priorities plus your invoice.
Every tool I use in ministry gets bent to serve the message, never the reverse. When a platform's defaults start pushing me toward content that performs instead of content that helps, I change the setting or I change the platform.
You will inherit systems somebody else designed, and most of your peers will treat those systems like weather. Ask who built this, what were they optimizing for, and who benefits from it staying this way. That question separates the people who run things from the people things get run on.
My kids are going to be handed tools built by people who have never met them and never will. So I am raising them to ask what a thing is doing to them, not only what it does for them. In our house the phone, the feed, and the assistant all get audited out loud.
This is book 3 of 9. These are books JP Morgan wants their new people and interns to read. What did you get out of going through this? Let's discuss this in the comments.
I love you.
This is the YEAR OF DISCOVERY.
Rick