I've made a lot of slide decks. Probably a few thousand over ten years, if you count the ones I deleted out of shame. So when AI presentation tools started showing up everywhere, I had two reactions at once: genuine curiosity, and a strong suspicion that this was another "we added AI" feature nobody asked for.

I've now used them enough to have an actual opinion. It's messier than either the demo videos or the skeptics suggest.

The first deck I let AI build

I gave it a rough prompt — internal product review, three features, one quarter of numbers — and got back fourteen slides in about forty seconds. They were competent. Clean layout, consistent fonts, logical order. Also completely hollow.

One slide said "Improve user engagement through better onboarding." That's not a finding. That's a note to self. The tool had no idea what our onboarding actually does, what we tried last quarter, or why the numbers moved. It just knew what a slide about onboarding usually looks like.

I spent about ninety minutes fixing it. Which is roughly what building from scratch would have taken, except I also had to fight the feeling that I was being ungrateful to the future.

There's a related trap I fell into. The decks look good. The layouts are clean, the spacing is right, the icons are tasteful. And a beautiful deck with nothing in it is worse than an ugly deck with a point, because the polish makes you trust it. I've sat in meetings where everyone nodded along to slides that were essentially generated filler, including me.

Where it actually earned its place

Two things changed my mind, neither of them dramatic.

The first is the blank-page problem. Going from nothing to a rough outline is the worst part of making a deck, and AI is genuinely good at it. Not because the outline is right, but because having a wrong outline is easier to fix than having no outline. I can argue with a bad structure. I can't argue with a cursor.

The second is cleanup. Reformatting, splitting slides that got too dense, turning a wall of notes into something readable. My teammate told me last week she pasted forty pages of user research into a tool and got a workable skeleton in about three minutes. Then she spent two hours rewriting it. So the honest math is two hours instead of a full day. Real savings, but not the "make your deck in five minutes" pitch.

Things I've handed off without regret:

  • Turning a page of messy notes into a first-draft outline
  • Rewriting titles so they say something instead of "Q3 Update"
  • Cutting a 30-slide deck down to 12 when I can't see what's redundant anymore
  • Translating a deck into Chinese without it reading like a product manual

Notice what's missing from that list. Any actual thinking about what we should do.

What I've settled on

My current approach is boring. I write the story myself, usually as a messy list of sentences in a note app. Then I use a tool to turn that into slides, and I fix everything afterward.

I think that division of labor is roughly right for now, though I'm not confident it stays that way. While building Huanjian, the argument my team keeps having is about exactly this line — how much of the thinking should the tool do, and how much should stay with the person. Nobody has won that argument yet.

What I keep coming back to is that presentations aren't really about slides. They're about deciding what to say to a specific group of people and why. AI is fast at the slides. It doesn't have a view on the why, and I'm honestly not sure it should.

So, no verdict. I use it most weeks. It saves me time on the parts I don't care about. The part I do care about is still mine, and I think I'm okay with that.