I want to start with an admission: for the first three months of my AI experiment, I felt worse. More anxious. More distracted. I was adding tools faster than I was learning to use them, and the cognitive overhead of managing six different assistants was quietly eating the time I thought I was saving.
If that sounds familiar, this post is for you. Not because I have a perfect system — I don't — but because I eventually found something that feels sustainable, and I want to walk through the specific choices that got me there.
The problem with "AI-first" advice
Most productivity content tells you to start with the tool and build a workflow around it. I did that. It's backwards. The only thing that made my setup click was writing down what I actually wanted my days to look like and then working backwards to what that required.
The question isn't "what can this AI do?" It's "what's the bottleneck in my current process, and is AI genuinely the right lever for it?"
For me, the bottleneck was first-draft paralysis. I can research and edit fluently; I have always found the blank page disproportionately expensive. That one insight — boring as it sounds — changed everything, because it meant I only needed one AI tool to do one thing well, not six tools doing six half-useful things.
What my stack actually looks like
After a year of experimentation, my daily workflow involves exactly four tools. I'm not going to recommend them by name because the landscape changes monthly, but I will describe the function each serves:
-
✏️
A drafting assistant — lives inside my writing app, used exclusively for generating rough first drafts from structured notes. I never publish its output directly; it breaks the blank-page problem.
-
📚
A research summariser — I paste in PDFs and raw links; it gives me structured summaries with citations I can verify. Cuts my research prep from three hours to about forty minutes.
-
📅
A calendar/task integrator — a simple automation that converts my unstructured voice notes into calendar blocks. Not glamorous, not impressive. Saves me 20 minutes every morning.
-
🔁
A publishing pipeline — auto-formats my markdown, generates SEO meta from the post body, and queues social snippets. Completely invisible once it's set up.
What I cut
I stopped using an AI meeting notetaker (it made me lazier in meetings), an AI email writer (it destroyed my voice), and an AI image generator for blog headers (took longer to prompt than to source a photo). The ROI on all three was negative once I was honest with myself.
The things that didn't change
My best work still comes from long stretches of uninterrupted thinking. AI didn't change that; if anything it made it clearer. When I use a drafting assistant as a crutch to avoid thinking, the result is mediocre and I know it before I finish reading it back. The tool only helps when I arrive with something real to say.
Reading widely, taking notes by hand, talking to people who disagree with me — these remain the actual inputs to anything worth writing. AI is useful in the middle of the process, not at either end.
If I could go back and tell myself one thing it would be this: the year you spend testing every new tool is not wasted. It builds a genuine intuition for what this technology actually is versus what the marketing says it is. That intuition is worth more than any individual productivity gain.
Take the slow road. Delete things. Notice what you miss and what you don't.