I got curious. And a little annoyed.
Everyone kept telling me AI would “10x my productivity.” LinkedIn was full of people claiming they replaced entire teams with ChatGPT. But when I actually tried it, my outputs felt… off. Generic. Like I was fighting the tool more than using it.
So I ran an experiment. For 30 days, I’d use AI for everything I reasonably could—every email, every document, every research task. If AI could help, I’d try it. And I’d track everything.
Here’s what actually happened.
The Setup
Before diving in, here’s how I structured this:
- Try AI first. Before doing any task manually, at least attempt it with AI.
- Track everything. Time spent, quality of output, whether I actually used the result.
- Be honest. If AI made things worse, I’d record that too.
- Use real work. Actual client emails, actual projects—no toy examples.
I mostly used Claude and ChatGPT, switching based on task type. (More on that later.)
Week 1: The Honeymoon Phase
Tasks attempted with AI: 47 Tasks where AI helped: 38 Tasks where AI made things worse: 3 Time saved: ~4 hours
What Clicked Immediately
Email drafting was the obvious winner. I’d write a rough brain-dump, paste it in, and get back something sendable in seconds. I probably handle 20-30 emails a day. AI cut my drafting time by at least 60%.
Meeting prep surprised me. Before calls, I’d paste in context about the person or company and ask for talking points, potential questions they might ask, things to research. Way more useful than I expected.
Summarizing long documents was a revelation. I had a 40-page contract to review. Asked AI to summarize key terms and flag anything unusual. It caught two things I would’ve missed—including a weird IP clause buried on page 27.
What Didn’t Work
Writing social posts. Output was generic. Felt fake. I’d spend more time editing than if I’d just written from scratch.
Creative brainstorming. Asked for product name ideas. The suggestions were either boring (“TaskFlow,” “WorkHub”) or trying too hard (“SynergiMax”). Not helpful.
Anything requiring my specific voice. AI doesn’t know how I talk. First drafts needed heavy rework.
Week 1 Verdict
Real gains, but concentrated in specific areas. AI excelled at structured tasks—emails, summaries, prep. It struggled with anything requiring creativity or personal style.
Week 2: Finding the Edges
Tasks attempted with AI: 52 Tasks where AI helped: 41 Tasks where AI made things worse: 5 Time saved: ~5 hours
New Discoveries
Code debugging was a game-changer. I’m not a developer, but I manage a small site. When something broke, I’d paste the error into Claude and get an explanation plus fix. Saved me from two-hour Stack Overflow rabbit holes.
Learning new concepts. Needed to understand unit economics for a project. Instead of reading five articles, I asked AI to explain it like I’m smart but new to the topic, with SaaS examples. 10 minutes instead of 2 hours.
Editing my own writing. Not first drafts—AI still struggled there. But editing? I’d write something, then ask “make this 30% shorter without losing meaning.” The suggested cuts were almost always good.
The Failures
Anything with emotional nuance. Asked AI to help craft a sensitive message to a team member. Technically correct. Emotionally tone-deaf. Had to rewrite completely.
Fact-checking. AI confidently told me a company was founded in 2018. It was founded in 2015. I now verify anything factual.
Complex analysis. Asked for competitive analysis of a market I know well. The output was surface-level and missed key players. AI doesn’t have insider knowledge.
Week 3: Building Systems
Tasks attempted with AI: 61 Tasks where AI helped: 52 Tasks where AI made things worse: 2 Time saved: ~7 hours
The Breakthrough
I realized I was wasting time re-prompting for similar tasks. So I built systems:
Email templates with variables. Created a prompt for each email type—intro, follow-up, proposal. Now I just fill in the blanks.
Standard prep docs. Before any meeting, same prompt: background, talking points, risks, questions. Saved as a shortcut.
Content outline generator. For any piece I need to write, a prompt that generates structure. I fill in the substance.
This changed everything. Instead of crafting prompts from scratch, I’m loading pre-built ones. Much faster.
Unexpected Wins
Meal planning. Gave AI my dietary preferences and asked for a week of dinners. Simple. Worked perfectly. Not sure why I didn’t think of this earlier.
Travel planning. “Build a 3-day itinerary for Lisbon focused on food and architecture. Mix touristy and local spots.” Better output than most travel blogs.
Decision frameworks. Stuck on something? AI walks me through a decision matrix without me having to set one up. Genuinely clarifying.
What I Stopped Using AI For
Gave up on creative writing entirely. AI’s creative output is mediocre at best, and editing it to not sound like AI took longer than writing myself.
Also started being more thoughtful about what I paste in. Not paranoid—just aware.
Week 4: The New Normal
Tasks attempted with AI: 58 Tasks where AI helped: 51 Tasks where AI made things worse: 1 Time saved: ~6 hours
By week 4, AI use became automatic. I didn’t think “should I use AI for this?"—I just did, for certain task types.
Tasks I now always use AI for:
- Email drafting and editing
- Summarizing long content
- Meeting prep and follow-up
- Explaining concepts I don’t understand
- Initial research (followed by verification)
- Data formatting and cleanup
- First-pass editing of my own writing
Tasks I learned AI can’t help with:
- Anything requiring my specific voice
- Creative work that needs to feel original
- Sensitive interpersonal communication
- Deep analysis in areas I know better than AI
- Anything requiring current or verified facts
My Numbers vs. The Research
Here’s what I tracked:
Total tasks attempted with AI: 218 Tasks where AI helped: 182 (83%) Tasks where AI hurt: 11 (5%) Tasks where AI was neutral: 25 (12%)
Estimated time saved: ~22 hours over 30 days That’s roughly 45 minutes per day.
Not the 10x productivity the hype promised. But meaningful.
What surprised me: my numbers actually match the research. A 2025 St. Louis Fed study found workers who use generative AI save 5.4% of their work hours—about 2.2 hours per week, or ~26 minutes daily on average. Frequent users saved 9+ hours weekly.
Anthropic’s research found Claude speeds up individual tasks by about 80%—tasks that would take 90 minutes without AI take around 18 minutes with it.
But here’s the thing that made me feel less crazy: a METR study of experienced open-source developers found they actually took 19% longer using AI tools—even though they thought they were 20% faster. The productivity gains depend heavily on the task type and how well you’ve learned to use the tools.
My 45 minutes daily is ~275 hours per year. That’s real time I can spend on work that actually requires a human.
The Honest Assessment
AI is great for:
- Structured output — Emails, summaries, outlines, templates
- Explaining things — Better than Googling for most concepts
- First drafts of routine content — Kills blank-page anxiety
- Processing information — Reading, extracting, reformatting
- Learning — Ask follow-ups, get instant explanations
AI is mediocre for:
- Anything creative — It’s derivative by nature
- Anything personal — Doesn’t know your voice, context, relationships
- Analysis in areas you know well — You’ll spot the gaps immediately
AI is actively bad for:
- Replacing thinking — It can assist decisions, not make them
- Emotional communication — Technically correct, humanly wrong
- Anything needing verification — Always check facts
What Stuck After 30 Days
These became permanent parts of my workflow:
| Use Case | Tool | Frequency |
|---|---|---|
| Email polishing | Professional Email Writer | Daily |
| Document summary | Executive Summary Generator | 3-4x/week |
| Meeting prep | Custom prompt | Before every call |
| Concept learning | AI Tutor | Weekly |
| Writing editing | “Make this 30% shorter” prompt | Every time I write |
| Decision frameworks | Decision Matrix Creator | Monthly |
Not revolutionary. But consistently useful.
What I’d Tell Myself on Day 1
Start with email. Easiest win. You’ll see results immediately.
Build prompts, not habits. Don’t try to “learn prompt engineering.” Just save prompts that work and reuse them.
AI is a drafter, you’re the editor. Never send or publish AI output without reading it.
Verify facts. AI is confident even when wrong.
It’s a tool, not a replacement. AI handles drudgery. It doesn’t replace judgment.
The savings compound. 5 minutes saved on one email is nothing. 5 minutes saved on 10 emails a day for a year is a lot.
Try Your Own Experiment
You don’t need 30 days. Try one week:
- Use AI for every email you send
- Summarize one long document
- Prep for one meeting with AI
- Ask AI to explain something that’s been confusing you
Track what works. Build your own list of keepers.
The goal isn’t to use AI for everything. It’s to figure out where it actually helps your work.
The Skills I Used Most
These five covered most of my use cases:
- Professional Email Writer — Easily #1
- Executive Summary Generator — For every long doc
- AI Tutor — Learning new concepts fast
- Meeting Notes Action Extractor — Post-meeting clarity
- System Prompt Architect — Building custom assistants
If you’re starting from zero, those five will handle 80% of what you need.
Ready to try your own experiment? Browse all skills or start with the Lazy Person’s Guide for the essentials.