Making presentations used to be my most avoided task. As a product manager I'd rather write a PRD or debug a workflow than slot content into slides. So when the current wave of generative AI hit, the first thing I tested was whether it could take over the presentation layer. After several months of daily use across different tools, here's what the experience actually looks like once you get past the novelty.
The middle stage is where AI earns its keep
The hardest part of any presentation isn't the slides. It's the blank space before you have a narrative. I've spent hours shuffling ideas around at the start of a deck because everything felt equally important.
AI solved that for me faster than I expected. I'll start a conversation with a rough list of points, even if it's just a few vague keywords. Then I ask for a narrative structure back: what's the story, what gets cut, what needs to be in the appendix instead of the main flow. What's useful is the articulation of why the story should fit together that way.
Recently I needed to pitch an internal efficiency idea to stakeholders who had very different priorities. My initial outline was just "cost reduction" over and over. The AI pushed back with a structure that made different kinds of value explicit at different points in the narrative. It took a few rounds to get there, but it did the intellectually boring work of rearranging my thinking so I could see where it was shallow.
While building Huanjian I've found that getting this first draft framework right can save more time than any amount of visual polish. Once the structure holds up, my job shifts from deciding what to say to reacting to what's been proposed. That dialectic is surprisingly effective.
The bright shiny mistakes that slip through
It would be a better story if I could report flawless results across the board. I can't. AI tools love confident-sounding outputs that are subtly wrong. Not factually wrong in an obvious way, but contextually wrong in a way that will make you sound silly in front of your audience.
Last month I asked a tool to write a slide about our pricing philosophy. It delivered four bullet points that sounded perfectly reasonable until I realized they described a freemium model. We're B2B. Free tier doesn't exist and never will. The AI had just pattern-matched "pricing philosophy" to the most common SaaS template in its training data.
That kind of error isn't hard to catch if you're actually familiar with the material. The problem is that it's persuasive at a glance. I've learned to read every generated line as if my smartest competitor wrote it, looking for places where wording masks an underlying assumption. That effort sometimes feels like double the work. But I also can't go back to writing everything from scratch.
What I've stopped doing is trusting AI for visual judgment. It will happily suggest a pastel color scheme for a financial report or a playful font for a serious compliance update. I don't think the current generation has any real understanding of what your audience needs to feel at a given moment. That's still my call.
The workflow that sticks
My honest process right now is hybrid. I use AI for the parts that involve transforming rough ideas into something I can react to. That includes the early narrative, a lot of the speaker notes, and the first round of getting my thoughts into organized form. I do not use it as an autonomous system that generates a complete deck without my supervision.
Every tool I've tried still produces slides where I want to strip out most of what's on the page and let the material breathe. Visually, AI tools have gotten better but it hasn't stopped being a collaborative process. When I'm genuinely pressed for time, I'll settle for an AI deck that's seventy percent there, but I never ship something that hasn't been through a manual pass with a critical eye.
The most honest summary is that AI has changed the texture of my work, not removed it. I still do the same amount of thinking. What's shifted is that the AI becomes a kind of draft partner who's always available to bounce an idea off. That's useful in a way I didn't anticipate. The output still needs to be wrangled, reorganized, and rewritten in enough places that I understand why the process doesn't feel entirely like magic yet. And maybe that's okay. The gap between where it is and where it needs to be is where the actual craft lives.