My new AI assistant handed me the three most important things to do that day.
I’d already done all three.
Confidently wrong, all three. Not a dramatic failure, just a quiet tell that the system still had gaps. A connection I hadn’t made yet. A question it had never thought to ask me. Something not yet tuned.
I should explain what it even is. This month I built my AI assistant. Not an app I downloaded, but an AI project space I gave a memory and a standing set of instructions, then told to watch my whole operation and tell me where to look. A chief of staff, basically. On its first real morning, it read a slightly stale picture of my world and told me to do work I’d already finished.
Which sounds like a story about a machine being dumb. It isn’t. It’s a story about the most human mistake there is, and what it cost me to keep making it.
The thing I’m good at, and the trap inside it
I’m good at building systems. Always have been. Give me a messy process and I’ll give you a clean one. Folders, routines, checklists, the works.
Here’s what took me decades before AI to recognize: nearly every system I’ve ever built quietly depended on me to stay alive. The checklist only worked because I remembered to open it. The routine only held because I was the one holding it. The whole operation lived in one place, and that place was my head (or my storage drives).
We have a flattering word for this. We call it being indispensable. We say it like it’s a trophy.
It isn’t. Indispensable usually just means it is not documented. It means the system would fall over the moment you stepped away, which is another way of saying you can never really step away. That’s not job security. That’s a cage with your name engraved on it.
I didn’t want to be the load-bearing wall of my own life anymore. So this month I tried to build my way out.
What I actually set out to build
I’m building a business, An Intentional Age, while actively working with clients as a fractional AI advisor. The pieces weren’t scattered across drawers and folders. They were organized, each living in its own Claude project: content in one, research in another, client work in a third. Everything had a home. What none of them had was a collective view. Even inside the same platform, the only thing connecting them was me, waking up every morning already carrying the whole picture.
So I wrote a job description for an assistant that could carry it instead. Think of it the way a busy executive thinks of a chief of staff: not someone who does one task well, but someone who watches the whole operation and tells you where to look. In my case, that “someone” is an AI I’ve been training on how to think and what to watch for.
I gave it three rules.
First, life before work. Before it says a word about business, it has to surface what matters more. My relationships. My health. My energy. In that order, because that’s the real order, even on the mornings I forget it.
Second, never just show me data. Anybody can build a dashboard that displays numbers. I wanted something that reads the numbers and tells me what they mean, then hands me the decision. Personalized synthesis, not a status board.
Third, it has to get sharper over time. A tool you have to reexplain every week isn’t a colleague. It’s a chore.
Simple enough to write on a napkin. Building right was another matter.
The lesson hiding inside the failure
Back to that first morning.
The reason it told me to do three things I’d already finished wasn’t that the AI was dumb. It simply hadn’t caught up to my current reality. A couple of Kanban project cards hadn’t been moved to “done” yet. An email it flagged as needing a reply had already been answered, but it judged the thread by when the message arrived, not by what I’d done since.
It was looking at a snapshot of time that wasn’t the immediate present, yet speaking with total confidence about the morning’s work.
And here’s the part that stopped me cold. That’s not a machine problem. That’s the most human thing in the world.
How often am I certain about a situation while looking at an outdated picture of it? Sure that a friend is still angry, when they let it go days ago. Convinced a plan won’t work, based on how things stood a month back. Confidence and accuracy feel identical from the inside. They are not the same thing.
So the founding lesson of the whole project turned out to be this: a system that synthesizes is only as good as what it can see. Catching it being confidently wrong isn’t the system failing. That’s the system working. That’s the whole point of having one you can check.
The exciting part was the boring part
I want to be honest about the middle, because the temptation is to tell you I built a robot and now I’m free.
Two of the three weeks were not clever features. They were cleanup. Real human knowing the real deal.
Merging three overlapping to-do routines into one. Untangling files that had quietly multiplied into four near-identical copies, so the system would stop tripping over which version of “me” was the real one. Settling on a single rule for the whole mess: who I am and how I sound lives in exactly one place, and everything else points to it. Never copied. Pointed to.
It was unglamorous. It was also the actual work.
That maps onto something I keep relearning about a life, not just a system. The bold third act, the reinvention, the big swing, none of it holds without the dull infrastructure underneath. The sleep you protect. The walk you take whether or not you feel like it. The relationships you tend before you need them. We want the breakthrough. The maintenance is the breakthrough. The exciting part really is the boring part, and pretending otherwise is how good intentions quietly fall apart.
Teaching it the one thing it can’t generate
The shift from clever tool to true assistant came down to two changes, and neither was technical.
First, I gave it a standard it isn’t allowed to drop below. If something it produces isn’t good enough, it doesn’t ship it to me. It hands back the fixes first.
Second, and this is the one that matters: when I correct it, the correction sticks. If I override its judgment, that lesson gets written down somewhere it reads before its next run. So my corrections compound instead of evaporating the moment the session ends.
Here’s what I’m actually teaching it. Not facts. Judgment. Which answer is wrong even though it looks right. Which client will balk at that phrasing. Which “efficient” plan blows up in month three. That knowledge didn’t come from a course. It came from thirty-five years of watching what good looks like and, more often, what wrong feels like a half-second before it happens.
That is the asset. And it’s the exact thing the machine can’t manufacture on its own.
There’s a quiet fear running underneath a lot of the AI conversation for people my age. The fear that decades of experience are about to be made worthless by a clever piece of software. I’d flip it. The experience is the part the software is starving for. Pattern recognition, the feel for context, the judgment that only thickens with the years. That doesn’t shrink as we age. It’s the one thing that grows.
Stanford’s Center on Longevity backs this up with actual numbers. The kind of knowledge that comes from accrued experience, vocabulary, judgment, the ability to read a situation, keeps rising through your sixties. Vocabulary scores now peak in the late sixties or early seventies. Older research assumed that peak came in your forties. It didn’t. The tests were wrong, not the brains.
You’re not the person AI replaces. You’re the person who finally has something worth handing it.
You give it your judgment. It carries that judgment into the hours you’re not there.
A home and a heartbeat
The last pieces were the least glamorous and, it turns out, the ones that made the whole thing real.
I gave the system a logbook of every change to how it works, so I can see what shifted and when. And I gave it a short list of its own fragile parts to check, so it can confirm nothing quietly broke since the last time it ran. It now runs on its own early Monday, before my protected morning hours, and lands one recommendation with a simple yes-or-no ask. I don’t have to go get it. It comes to me.
People build the dashboard and stop. The pretty screen that shows you everything. That’s why most of these systems die on the first tired morning: a dashboard still needs you to open it, read it, and decide. I needed three things instead. A brain that remembers across sessions. A face that hands me one decision, not a wall of data. And a heartbeat, so it runs and improves without me starting it.
Brain, face, heartbeat. Skip any one and the whole thing collapses back onto you, which is right where you started.
What it actually gave back
Here’s the payoff, and it isn’t the one I expected.
I thought I was buying time. Saving an hour here and there. That’s real, but it’s not the thing.
The thing is that I get to wake up as a person again. For years I opened my eyes already running the operation, the whole list arriving before my feet hit the floor, the people right in front of me getting whatever attention was left after the machinery in my head took its cut.
Last week Julie was pulling together a grandparents camp week for the kids, a run of days with specific events, one of them down at the tide flats here in Tacoma. She was checking whether I could make it.
I’m the one who caught the conflict, not her, and only because I happened to check the calendar myself in time. A commitment I’d made weeks earlier was sitting quietly on a calendar my new chief of staff had never been told to look at.
It hadn’t caught it either. Not because it was careless, but because I’d never connected it to the one calendar that actually mattered that week: the shared one with Julie. It was reading a version of my life that stopped at the edge of the business. My own blind spot, built into a system meant to catch my blind spots.
I moved things around in time. The tide flats stayed on the calendar. What almost slipped wasn’t a client call. It was an afternoon with my grandson, who is attuned enough to his “gan-pa” that he’d have felt it if I hadn’t shown up. So the fix went where every other lesson goes now: connect it to our shared calendar too, not just mine.
That’s not proof the system is finished. It’s proof of what’s different now. A month ago, a miss like that wouldn’t have almost slipped. It would have just slipped, quietly, and I wouldn’t have felt the cost until I was standing at the water without him.
When the operation no longer lives only in me, I can show up to the people I love not already depleted. That’s the actual return on three weeks of file cleanup. Not a tidier business. A less divided attention.
You’ve got your own version of this. The person whose Tuesday almost got missed. The call you almost forgot. The moment that almost slipped past you too, before something outside your own head caught it in time.
The one thing to do this week
You don’t need three weeks. You don’t need to build any of this.
Here’s the small version, and it’s the part that actually counts: find the one thing that lives only in your head. The standing worry you carry. The weekly task only you know how to run. The piece of knowledge the whole thing depends on. Pick one. Write it down somewhere outside yourself this week. A note, a doc, a card on a board. Or open whatever AI tool you already have and just tell it. Not a prompt, not a project. One sentence: here’s the thing only I know, remember it for me.
Then notice what it frees up. That small relief is the whole idea, scaled down to something you can do before lunch.
What’s the one system in your life that would fall over if you stepped away? And what would it give back to you if it didn’t have to lean on you to stay standing?