Six months ago I made a weird bet with myself. I would use an AI tool to generate every single presentation I gave at work — from quick status updates to quarterly reviews to client pitches. No slide handcrafting, no manual layout tweaks. Just prompt, edit, deliver.

I'm a product manager, not a designer. Presentations have always been a necessary evil. I figured AI could save me the drudgery. And it did, but not in the way I expected.

The speed trap

The first month was intoxicating. I'd type a rough outline, and within 30 seconds I had a 15-slide deck with charts, bullet points, and a consistent theme. I cut my prep time by more than half. I started saying yes to more meetings because the presentation cost felt near zero.

But then I delivered one of those AI-generated decks to the engineering team. Midway through a slide on "Architecture Decisions," someone raised their hand. "This chart doesn't match our data." He was right. The AI had invented numbers that looked plausible but were completely wrong. I'd skimmed the output, trusted the pretty graph, and missed it.

That's when I realized the time I saved on slide generation got spent on fact-checking, rewriting, and rebuilding logic. The net gain was maybe 10%, not 50%. And the mental fatigue was different — instead of wrestling with layout, I was wrestling with hallucinations.

This is something we've thought about while building Huanjian. The easy win is making slides fast. The hard win is making them accurate and yours.

Where my voice went

About three months in, a colleague told me something uncomfortable. "Your presentations feel robotic. Like someone else wrote them." She was right. I'd been editing for correctness but not for personality. The AI defaulted to a bland corporate tone — lots of "leverage synergies" and "drive transformation." I hate those phrases, but I stopped catching them because the slides looked so clean.

I tried to fix this by adding my own anecdotes and quirks in the speaker notes. But the visual structure of the deck still pushed a formal narrative. My team started asking if I was outsourcing the work. I wasn't — but I'd outsourced the voice.

I learned to force a "human rewrite pass" where I'd actually delete whole slides and rebuild them from a single sentence in my own words. Awkward sentences, half-finished thoughts, inside jokes — those made the presentation feel alive again. The AI was good at scaffolding, but terrible at soul.

The things AI did well

I'm not going to trash the experiment entirely. There were three things AI genuinely made better.

One: structure. When I'm stuck on how to tell a story, feeding a raw list of facts into an AI gives me three or four different outlines. I pick one, tweak it, and move on. That alone saves half an hour per deck.

Two: data exploration. I once had a messy spreadsheet of user session data. I asked the AI to find the most interesting three trends and turn them into slides. It wasn't perfectly accurate, but it gave me a starting point I wouldn't have found on my own. I then verified everything manually.

Three: the boring repeaters. Monthly status reports, weekly sprint reviews — those are templates. AI can fill them in without me re-typing the same bullet formats. The risk is that they all start to look the same, but for internal meetings that's often fine.

What I do now

After six months, I still use AI for almost every presentation. But the process has changed. I treat the AI like a junior designer with a bad habit of making stuff up. I give it constraints, I review every number, and I rewrite until the voice sounds like me. It's not a magic pill — it's a tool that amplifies what I put in, for better and worse.

The biggest surprise was how much I missed the friction of building slides by hand. That friction forced me to think about what I actually wanted to say. Now I have to build that friction back deliberately. Weird lesson from a productivity experiment: sometimes speed makes you slower where it counts.

I'm still figuring out the right balance. If you've tried this yourself, I'd honestly like to know what worked for you. Because I'm not sure my approach is final yet.