Do we really need AI to tell us how to stand in front of a room and talk? Yes, but only because the current crop of “AI slide generators” are essentially just fancy, skin-deep template fillers that ignore the actual human act of presenting.
For the last two years, the AI productivity space has been obsessed with the artifact. You feed a prompt into a tool, and it spits out ten slides with a vaguely cohesive color palette and some bullet points that sound like they were written by a corporate HR bot. It’s a visual trick designed to make the user feel like the work is done, when in reality, the hardest part of the job hasn’t even started. The problem is that a visually plausible deck is not a presentation; it’s just a document you project onto a wall. It is like buying a professional-grade kitchen but having absolutely no idea how to sauté a scallop—you have the equipment, and the room looks great, but the output is still a disaster because you’re missing the technique. We have spent an enormous amount of compute power on the aesthetics of the slide, while the actual communication remains a black box of anxiety and awkward pauses.
DeepSlide, detailed in a recent arXiv paper, tries to move the goalposts. Instead of just optimizing for the file, it looks at the delivery—the pacing, the narrative arc, and the actual preparation process. It treats the slide deck as a supporting actor rather than the lead. The focus shifts from “what does the slide look like” to “how does the speaker move through the information” (probably because most of us are terrified of public speaking). By treating the presentation as a performance rather than a static document, it attempts to bridge the gap between the digital asset and the human voice. It is a pivot from graphic design to orchestration, focusing on the temporal element of the pitch rather than the spatial layout of the page.
The industry has spent way too much time on the generation phase. We have reached a point of diminishing returns with hallucinated bullet points and AI-generated stock photos of people shaking hands in sterile offices. The real bottleneck isn’t the creation of the slides—it’s the gap between the PDF and the performance. If a tool can’t tell a speaker when to pause for effect, how to modulate their tone, or how to bridge two disparate ideas without sounding like a robot, it’s just another toy for people who want to avoid doing their own work. We don’t need more content; we need better delivery. The obsession with “one-click” generation is a distraction from the fact that most people are still terrible at explaining their ideas once the slides are actually on the screen.
Of course, the real-world friction here is immense. Imagine the latency of a real-time pacing assistant attempting to analyze your voice in a noisy conference room, or the sheer social cringe of a machine whispering in your ear while you’re trying to convince a board of directors to increase your budget. There is also the technical overhead—running a low-latency audio-analysis loop while simultaneously managing a high-res slide deck is a recipe for a system crash at the worst possible moment. Who actually wants a digital coach interrupting their flow in the middle of a high-stakes pitch? Or maybe not—maybe we’re so desperate for a coherent narrative that we’ll tolerate the awkwardness of a machine telling us we’re rambling. The friction isn’t just technical; it’s psychological. There is a very thin line between “helpful pacing” and “having a machine tell you that you’re talking too fast” while you’re already sweating under a spotlight.
The artifact is the easy part.
By Q4, we will see the major office productivity suites stop bragging about their “auto-slide” features and start integrating delivery coaching and narrative pacing tools based on this shift in focus. The novelty of the “one-click deck” has worn off, and the market is starting to realize that a pretty slide deck is useless if the person presenting it is incoherent. The next fight is over who can actually make the human standing in front of those slides look competent.
