Saturday morning I sat down to clear a pile of work I'd been punting all week, and I couldn't do it. Not wouldn't.. couldn't. I'd deliberately shoved a handful of decisions to the weekend because there was no room for them Monday through Friday, and by the time the weekend actually showed up my brain wasn't in a place to pick any of it back up.
Meanwhile the machine I walked you through last week had a perfect week.
Everything worked exactly like I said it would
Thirty-eight scheduled jobs fired sixty-nine times a day without me, and the 7am briefing hit my phone before I was out of bed every single morning. Agents wrote and reviewed changes to our own software overnight. The checkpoint that sits in front of every command blocked 868 things over the last thirty days and never once needed me to babysit it. The whole operation averaged about twenty-one bucks a day across roughly 1,400 unattended runs, and I didn't touch the machine.
So by every measure I put in front of you last week, it was a great week. And I ended it running on empty.
Those two facts sitting next to each other have been bugging me all weekend, because they're the actual story of what this shift does to you, and it's the part I keep not seeing anybody write about.
You can only automate the parts you can write down
Every single thing I've successfully handed to the system has one thing in common, which is that I was able to write it down. The 200-line file that reads more like a constitution than a prompt, the 139 written procedures, the 707 accumulated learnings.. every one of those exists because I could describe a job clearly enough that something else could go do it. That's the whole trick, and it's less impressive than it sounds. If you can articulate it, you can delegate it. If you can't, you're still holding it.
So over the last ten months I've systematically offloaded every part of my job I could put into words, which means what's left on my desk is, by definition, the stuff I couldn't.
And that stuff is a whole lot heavier per hour than what left.
What's actually left
Deciding what's genuinely worth building, when six things all look reasonable and only two of them matter. The people conversations that don't have a good version, only a less bad one. Telling a client something true that they don't want to hear, in a way that keeps the relationship. The judgment calls where being wrong costs real money and there's no test suite that tells you before you find out.
None of that got easier this year. There's just more of it per day now, because the easy stuff evaporated and the hard stuff didn't.
It's been getting clearer to me that this is just the job. All the worst problems go uphill.. every genuinely hard thing eventually lands on the person who can't hand it anywhere else, and you're the last stop. [Gallup's 2026 workplace report](https://www.gallup.com/workplace/349484/state-of-the-global-workplace.aspx) has leaders more likely than the people they manage to have felt stress, anger, sadness and loneliness the previous day, by 7 to 12 points depending on which one you pick, and I'd bet a lot of that gap is exactly this. Elon has that line about running a company being like eating glass and staring into the void, and it gets quoted like a flex. It never read like one to me. It's a fairly plain description of choosing to spend your days on the pile of things nobody else can carry.
Some weeks that's genuinely great. Then you sit down Saturday morning and the tank's just empty.
Automating your team's work can quietly make their week worse
This isn't a founder-feelings thing. It's an operating risk, and a lot of leaders are walking straight into it this year.
If you automate the legible work on a team and change nothing else about how that team runs, you haven't reduced anybody's load. You've concentrated it.
Whoever on your team was already carrying the judgment is now carrying only judgment, all day, without the small breather that formatting a deck or reconciling a spreadsheet used to quietly provide. That busywork was mostly garbage and I'm glad it's going.. but it was also recovery time, and not one of us put that on the ledger when we took it away.
I told you a couple weeks back that AI doesn't replace your team, it promotes them. I still believe that. But a promotion is also a heavier job, and if you hand somebody one without changing a single thing about the shape of their week, you're going to burn out your best person and then mistake it for a performance problem. That's the failure mode I'd be watching for, and it won't announce itself.. it'll look like your sharpest person getting quietly slower.
What I'm doing about it
I don't have this solved, so I'm not going to hand you a tidy framework.
What I've started doing is saying it out loud, because for months I kept score on what the system finished and didn’t really count what got pushed onto me as a result. If half my week is now decisions that can't be delegated, that's a real capacity number and I should be planning against it instead of acting surprised every Friday.
I've also stopped treating punting as a failure. Deciding not to do something this week is a legitimate output, and pretending otherwise is how the pile gets to a size where you can't pick any of it up.
And I'm trying to protect the recovery time on purpose now that the accidental version is gone. That one I'm genuinely bad at, and I'd be lying if I told you I've cracked it.
Wrapping up
The machine ran beautifully all week and I still hit Saturday with nothing left. You hand off everything you can describe, and in exchange you spend your whole day on the stuff you can't. Most days I'd take that deal again. Saturday I wouldn't have.
Anyway, that's where my head's at. If you've automated a bunch of your team's work this year and everybody seems fine, go check on whoever's carrying the judgment.
Putting this into action
Count what got pushed uphill, not just what got finished. For every workflow you automated this year, name who's now holding the part that couldn't be automated. If that's one person, you've built a single point of failure and put a nicer dashboard on top of it.
Audit for the recovery time you deleted. The low-stakes busywork your team lost was also their breather between hard things. Ask your sharpest person what their week feels like now versus a year ago, and actually listen to the answer.
What I'm reading
Polsia, "AI that runs your company while you sleep". That's their actual headline, and it's a nearly word-for-word description of what I spent last week's issue showing you.
Message your other Claude Code sessions. Your agents can now list and message each other across machines. Small feature, big shift.. this is the point where you stop running sessions and start running a team of them, which is a different management problem than a tooling one.
Marcus, an AI employee for Slack, Teams, email and SMS. The off-the-shelf version of what I built James to do. If you don't want to spend ten months building your own, this is the shape of the thing you'd buy instead.
anydoc. Converts Word, PowerPoint, Excel, PDF and the rest into clean Markdown. Deeply unglamorous and exactly the kind of plumbing that decides whether your AI can actually read your company's files.
Some AI tips for this week
If you can't write it down, you can't hand it off. Before you try to automate anything, try to describe the job in a paragraph. If you can't, that's not an AI problem and no model will fix it.. it means the process only exists in somebody's head, and that's the actual work.
Put a number on your undelegatable hours. Look at last week and mark every block that could only have been done by you. That number is your real capacity, and everything else you're saying yes to is coming out of it.
Onward & upward 🤘
Drew
P.s. If we haven't met yet, hello. I'm Drew Burdick, Founder and Managing Partner at StealthX. We work with brands to design and build great customer experiences that win. I share ideas weekly through this newsletter and over on the Building Great Experiences podcast. Have a question? Feel free to contact us. I'd love to hear from you.