For a stretch of years I ran a public agency. I was also building a small company, working a piece of land, and raising two boys. Every one of those fronts generated paper — email, filings, meeting notes, manuals, receipts, ideas — and for years, the paper won more days than I did.
The standard executive answer is delegation, and I delegated plenty. But there’s a category of information work you can’t hand off: the cross-domain stuff. The email that touches both the budget and a board relationship. The article that matters to the business you’re building, not the one you run. Nobody else can file that for you, because nobody else lives at the intersection.
So I built a system that does.
The part that took longest was admitting what the problem was
For about two years I thought I had a tools problem. I did not have a tools problem. I had tried most of the tools, some of them twice, and each one worked beautifully for a few weeks and then quietly became another place to not-look-at things.
The actual problem was that every one of them required me to show up. They were all vending machines: walk up, ask a question, get a thing, walk away. Nothing happened while I was in a four-hour budget meeting, which is exactly when the pile grows.
What I wanted was standing orders. Not a tool I operate — an operator that works whether or not I’m at the desk.
What that turned out to mean
A vault of more than seven thousand plain-text notes, organized by domain, sitting on my own machine. An agent that runs on a schedule against a written rulebook. Overnight it reads the intake channels, summarizes what arrived, links each item to what it relates to, and routes it to its home. By the time I pour coffee, the triage that used to eat the front of my morning is a short briefing I can read before the cup is empty.
The rulebook is the interesting half, and it’s written up properly here. The short version is that an unattended system operating on your life needs discipline far more than it needs intelligence, and every rule I gave it came from running operations rather than from software. Running an operation where people’s appointments depend on the process teaches you precisely what happens when a process fails silently. I applied the same standard to my own desk.
What I got wrong first
Two things, both embarrassing in hindsight, both instructive.
The first was letting it be confident. The early router filed everything, including the things it had no business being sure about, and I spent a month finding notes in reasonable-but-wrong places. The fix wasn’t a smarter classifier. It was giving the system permission to fail — anything it can’t confidently place goes to a queue for me. A system that admits uncertainty is worth more than one that’s usually right.
The second was celebrating. Every completed job announced itself, and within a week I had trained myself to ignore the notifications — which meant I was also ignoring the failures. Now success writes a log nobody reads and failure raises something I actually look at.
What it actually changed
Honestly? My mornings, and my attention.
The volume of information didn’t shrink. It grows every year. What changed is what it costs me. Filing costs me almost nothing now. Finding costs me almost nothing. What’s left is the part that was always the real job: judgment.
That’s the whole return. Not speed, not output, not any of the things this kind of project usually gets sold on. Just the removal of a tax I had been paying, daily, on the least valuable work I do.
■