On day three I caught myself watching a progress bar instead of working. That was the moment I realized something had shifted.
For a month I ran an experiment. Anything in my job that was repeatable, predictable, or mind-numbing, I handed to AI assistants first and only stepped in when they failed. Email triage. First drafts. Meeting notes. Research. Scheduling. The whole tedious layer that eats a workday without ever feeling like work.
I expected a productivity miracle. What I got was stranger and more useful than that.
Handing the busywork to AI assistants for 30 days gave me back roughly 8 to 10 hours a week, but only after I stopped treating them like magic and started treating them like junior staff. The wins were real on repeatable tasks. The losses were real on anything needing judgment. The biggest surprise was not how much time I saved, but how much it exposed about which parts of my job actually mattered.
I needed guardrails or this would just be me playing with toys.
The rule was simple: for any task, I had to try delegating it to an AI assistant before I did it myself. If the output was good enough, I shipped it. If it needed a fix, I noted the fix and moved on. If it failed twice, I took the task back and logged it as "not ready."
I tracked three things in a plain spreadsheet. Time spent setting up the task. Time spent fixing the output. Whether I'd trust it again next week. That third column turned out to be the only one that mattered.
The first few days felt incredible. I'd describe an email reply in one line and get something I could send with light edits. I summarized a 40-minute call in seconds. I generated a research brief while I made lunch.
Then the hangover hit.
I noticed I was spending real time reviewing everything, because I didn't yet trust it. A task that took me ten minutes now took eight minutes of AI plus six minutes of me checking it. That's worse. The first lesson of automation nobody mentions: in week one you pay a trust tax, and it's expensive.
The fix was not to check less. It was to delegate only the tasks where a small mistake was cheap. A rough draft can be wrong. A client invoice cannot. This trust tax is exactly the gap my honest review of which AI productivity tools actually earn their place keeps coming back to, and research like the Stanford HAI AI Index shows adoption running far ahead of measured, reliable returns.
Once I sorted tasks by the cost of being wrong, the trust tax shrank fast. On a cheap-mistake task, I could skim the output, fix the obvious, and move on without the anxious double-checking. On an expensive-mistake task, I either reviewed it carefully or didn't delegate it at all. That single sorting rule — how much does an error here cost me? — turned out to be the difference between week one feeling like a tax and week three feeling like a gift.
By week two I had a clear map. Here's the honest split.
| Task | Handed to AI | Verdict |
|---|---|---|
| First drafts of emails | Yes | Kept — huge time saver |
| Meeting summaries | Yes | Kept — better than my notes |
| Research roundups | Yes | Kept, with fact-checking |
| Scheduling and reminders | Yes | Kept — set and forget |
| Strategic decisions | Tried | Took it back |
| Anything with my actual opinion | Tried | Took it back fast |
The pattern was obvious once I saw it. AI assistants are brilliant at the shape of work and weak at the substance. They can produce a perfectly structured strategy memo that says nothing true. They cannot decide what your business should do next. They can absolutely write the email announcing the decision once you've made it.
This distinction became my filter for everything. Before handing off a task, I'd ask whether it needed my judgment or just my hands. Drafting a polite decline to a vendor? Hands. Deciding whether to fire that vendor? Judgment. Summarizing a report? Hands. Deciding what the report means for next quarter? Judgment. Once I sorted tasks this way, the delegation almost made itself. The hands-tasks went to AI without a second thought, and I stopped feeling guilty about it, because I was reserving my judgment for the things that actually deserved it.
The machine handles the form. You still own the meaning.
The single biggest win was boring, and that's why it worked.
Every morning I used to spend 45 minutes clearing my inbox. Sorting, replying to the easy ones, flagging the hard ones. I set up an email automation flow that drafted replies to the routine messages and grouped the rest by urgency.
Now my inbox is pre-sorted before I open it. The easy replies are drafted and waiting. I approve or tweak, and the genuinely important emails sit in a short list I actually read carefully.
That one change saved more time than every flashy AI demo combined. It's the same boring win I describe in how I cut twelve hours a week with three repeatable AI workflows. The unglamorous truth is that automation pays off most on the tasks too small to feel worth automating. We dream about AI doing something heroic, and meanwhile the real wins are hiding in the forty-five minutes of inbox sorting we never thought to question. The big, impressive automations are fun to talk about and rare to actually use. The tiny, boring ones run every single day.
Here's the part I didn't expect to write.
When AI absorbed the busywork, I had to face what was left. And a surprising amount of my "work" had been busywork all along. The reorganizing, the formatting, the looking-busy. Stripped away, my real job was maybe three hours of actual thinking a day, surrounded by motion.
That was uncomfortable. It's easy to feel productive when you're busy. It's harder when the busy is gone and you're left with the hard, slow, valuable thinking you'd been avoiding.
The 30 days didn't just save time. They showed me where I'd been hiding.
If I started the experiment over tomorrow, I'd change three things, because I wasted the first week learning lessons I can now hand to you for free.
First, I'd build a personal context document before delegating anything. The reason my early AI drafts felt generic was that the assistant didn't know my voice, my clients, or my non-negotiables. The day I wrote a one-page brief describing how I talk, who I serve, and what I never say, the output stopped sounding like a stranger wearing my name. Context is the difference between an assistant that drafts and an assistant that impersonates you well.
Second, I'd batch my reviews instead of checking everything in real time. In week one I interrupted my own focus a dozen times a day to approve tiny outputs. That constant context-switching cost me more than the AI saved. By week three I let drafts pile up and cleared them in two fixed review windows, morning and late afternoon. Same work, a fraction of the mental tax.
Third, I'd keep a running "took it back" list and revisit it monthly. Tasks that failed in January were suddenly viable by February as I got better at briefing and the models got sharper. The boundary between what AI can and can't do isn't fixed. It moves, usually in your favor, and only people who keep a list notice the moment a task crosses over.
There's a fourth lesson that's less practical and more honest. The experiment worked best when I stopped treating it as a competition with the machine and started treating it as a partnership. The framing of "can AI do my job" is a trap. The useful framing is "which 60% of my job is the machine better at, so I can spend my whole self on the other 40%." That reframe changed everything.
Delegation isn't surrender. It's deciding what's actually worth your hands.
If you want to run your own version of this experiment, start with one boring task this week and keep a simple log of what you'd trust again.
Q: Did you actually trust the AI to send things on its own? No, and I'd warn against it. Everything went through me before it left. The time savings came from drafting and sorting, not from removing the human entirely.
Q: Which AI assistant did you use? I rotated between a couple. The specific tool mattered less than the habit of delegating first. Any capable AI assistant with good context handling works.
Q: Didn't the constant reviewing cancel out the savings? In week one, almost. By week three, no. Once you learn which tasks are low-risk, you stop over-checking and the math flips hard in your favor.
Q: Would this work for a non-desk job? Less so. The more your work is information in, information out, the bigger the win. Hands-on work benefits far less.
Q: What's the one thing you'd tell someone starting this? Start with the most boring task you have, not the most impressive one. Boring tasks are predictable, and predictable is where AI shines.
Thirty days of letting AI assistants run my busywork taught me one thing I keep coming back to: the goal was never to do more, it was to do less of the wrong stuff.
The time I got back wasn't the prize. The clarity was. I now know which two hours of my day actually move things forward, and I protect them like they're made of gold.
If you tried this for a week, what would you discover was busywork all along? That's the question worth sitting with before you automate a single thing.
One person, output that looks like five. It isn't about working more hours — it's about a kind of leverage teams rarely have.
One idea a week to a published issue in under an hour. The boring system behind a newsletter I never dread sending.
Behind a lot of lean, profitable companies is the same small stack of AI tools. Here's what's actually running the show.
Comments
Sign in to join the conversation
No comments yet. Be the first to share your thoughts!