I Replaced 6 Hours of Weekly Busywork with AI. Here's Exactly What I Automated (and What I Didn't)
A real breakdown of which tasks AI actually handles well for productivity, which ones it makes worse, and how to set up a weekly AI workflow that saves time without the hype.
TL;DR
I tracked a week of my own work. AI tools saved me about six hours. But they also wasted two. The tasks AI handled best: routine email, document summaries, meeting prep, and spotting patterns in my own productivity data. The tasks AI made worse: calendar scheduling, creative brainstorming, and anything requiring emotional intelligence. Net gain: roughly 50 minutes of usable time back every day. The key was learning what NOT to hand off.
I tracked a week of my own work to find out. The result: AI tools saved me about six hours. But they also wasted two.
That ratio (6 hours saved, 2 lost) only happened after I learned what not to hand off. The first time I tried to use AI for everything, I probably broke even at best. The AI would generate something, I'd spend 20 minutes editing it, and I'd wonder whether I was actually ahead.
Here's what worked, what didn't, and the workflow I landed on.
Email that doesn't need your personality
The single biggest time save was email. Not the thoughtful ones. The routine stuff: scheduling confirmations, status updates to people who just need a status update, "here's the document you asked for."
I wrote a short Claude project prompt with my typical sign-offs, my preference for short sentences, and a rule to never use the word "circle back." It handles those in under 10 seconds each. I used to spend 15-20 minutes a day on them. That's an hour a week saved, and nobody noticed. Nobody was supposed to, that's the point.
Emails that require judgment or actual relationship maintenance stay with me. AI can't read the room in a thread where someone is frustrated or confused.
Summarizing things I should have read but didn't
I get a lot of PDFs. Research reports, product specs, meeting notes I wasn't in. The Claude project I use for email also handles summaries. I drop in a PDF, ask for a one-paragraph summary and three bullet points, and spend 90 seconds reading instead of 15 minutes skimming.
This saves around 90 minutes a week. The catch: if it's something I'm going to make a decision about, I read the original. The AI summary tells me whether it's worth reading the full thing. It doesn't replace the full thing.
Meeting prep that actually helps
Before important meetings, I paste the agenda and any prep docs into Claude. I ask for three questions I should probably ask but might not think of. Takes 2 minutes. About half the time, one of those questions turns out to be genuinely useful — something the room was avoiding or a connection nobody had made yet.
This isn't AI doing the thinking. It's AI being a second pair of eyes that isn't emotionally invested in the meeting's outcome. That detachment is useful.
Tasks where AI was actively worse
I tried using AI for calendar scheduling. It was bad. Claude and ChatGPT don't see my calendar, don't know my energy patterns, and can't read tone in scheduling emails to tell whether someone is being flexible or firm. I spent more time correcting the AI's scheduling attempts than I would have spent just doing it myself.
I also tried AI for brainstorming creative work. The ideas were fine. That was the problem. They were all fine, none were interesting, and after reading six perfectly reasonable suggestions I felt like my own thinking had gotten duller. I stopped. For creative work, I now use AI only after I've come up with my own ideas first — to stress-test them, not to generate them.
The workflow that stuck
After a month of experimenting, here's what my AI-assisted week looks like:
Monday morning: Dump the week's priorities into a Claude thread. Ask it to spot conflicts, flag things I probably forgot, and order tasks by dependency. Takes 5 minutes.
Daily: Process routine email (5 minutes). Summarize anything long I need to read (10 minutes spread across the day).
Before big meetings: Paste the agenda and any prep docs. Ask for three questions I should probably ask but might not think of. Takes 2 minutes.
Friday afternoon: List what got done, what didn't, and why. Ask the AI to spot patterns in what keeps getting delayed. I learned I consistently over-commit on Wednesdays. The AI noticed before I did.
Total time using AI tools: about 40 minutes a day. Time saved: about 90 minutes. Net: roughly 50 minutes of usable time back every day.
That's not life-changing. But it's real. And unlike most productivity advice, it didn't require changing my personality or waking up at 5am.
The rule that made the biggest difference
After weeks of trial and error, I landed on one rule that improved my AI productivity more than any prompt trick: never ask AI to do something you can't evaluate yourself.
If I can't tell whether the output is good, I don't hand it off. Period. This rule eliminated about 80% of my AI-related time waste. The remaining 20% is when I break my own rule because I'm tired and think "it'll probably be fine." It's never fine.
FAQ
Which AI tool is best for productivity?
Claude for writing, summaries, and analysis. ChatGPT for quick questions and code. Gemini for anything involving Google services. I use Claude for about 70% of my productivity workflow.
Will AI eventually handle calendar scheduling?
Probably, when it has full access to calendars, email, and enough context to understand priorities. But today's tools can't do it well. The integration isn't there yet.
How do I avoid spending more time managing AI than actually working?
Pick 2-3 specific tasks to automate. Use the same prompt structure every time. Don't optimize past "good enough." The people who lose time to AI are the ones constantly tweaking their prompts instead of shipping.