AI Deck Tools Are All Pointed the Wrong Way—What You Wanted Was Never an Editable PPTX

Let me lead with a call that may offend some people: I think the AI deck applications available today, every single one of them, have gone off in the wrong direction.

Why do I get to say that? Across seventeen or eighteen years of my career, an enormous amount of my time went into planning and strategy, and the main job was writing decks—everyday pitches, keynote scripts for launch events, a hundred to two hundred of them a year on average, not counting revisions. I even served as an internal PPT instructor at Tencent. So on this subject I probably have some standing.

Everyone’s competing over a point that doesn’t matter

What I’ve noticed is that today’s AI deck tools have all walked into the same big trap: they burn enormous time and resources on how to generate an “editable” deck—one you can drag with a mouse, where you can edit individual characters. Some even turn that into a selling point and push it as their differentiator.

I think it’s a total waste of time.

Back to the original question: why do we have decks at all

Think about why a file format like PPT exists in the first place. PPT stands for PowerPoint, and what it wants is to power the point: while you’re speaking, you need something visual to drive your core argument home so the audience follows you better. That’s all. Everything else exists to make that happen faster.

Which means this. As long as the points I want to make in a talk get visualized page by page, quickly, and revised the way I want when I need changes, that’s enough. Whether you can drag it with a mouse or edit individual characters—none of that matters.

To be straight with you, I haven’t opened deck software in about a year. All my outside teaching and all my pitches run as web pages, presented as slides in the browser. It works with a presentation clicker, and one link pushes a complete, crisp version to everyone in the room instantly.

How I make decks with AI now

Roughly four steps:

1. Have Claude Code interview me first. I dump whatever this pitch or class is meant to cover straight at it, without overthinking. It comes back with all kinds of questions to fill in what the deck needs—who the audience is, what they want and where their pain is, and what outcome I’m after.

2. Have AI run concurrent multi-agent research from several angles to fill the deck with substance. I also put out my own core arguments for this pitch along the way, and the narrative logic I want—I’m not handing AI the wheel entirely. But it stays light. One conversation and it’s done.

3. Have AI produce a black-on-white “big-type poster” web version first. I can open any page and see its information structure at a glance, along with the chapter breakdown of the whole talk and the thread of narrative logic running through it. Then I keep working each page’s content over with AI until the text hits its first locked version. AI revises web pages fast—minutes. And it doesn’t just change the text on each page to my specs; it will re-sequence the pages when I change my narrative approach.

4. Once the black-on-white version is confirmed and I’ve read it through myself, have AI use the GPT Image 2 API to generate the pure-image deck concurrently. It turns the black-on-white version into a finished product that’s genuinely beautiful, with image elements and even charts, all in one consistent template style. Every generated page is a single complete image, yet the style holds together, and the content follows the confirmed black-on-white version strictly. At this stage I can also give it template references (“look how well this deck is done”), or hand it the client’s logo and brand elements so the styling matches the client’s colors.

Five to seven days became fifteen minutes on the way to class

Do you know how long it used to take me to write a proposal like this? At the absolute fastest, five to seven days to finish a fifty-plus-page, finely typeset course deck.

The deck I’m showing here? The night before, I was still adjusting the black-on-white outline with the organizer. The next morning, on the way to the classroom, fifteen minutes was all it took to generate the whole thing. And it turned out well.

So were all those years of practice wasted

Sometimes I do get wistful: I honed this skill for all those years, I was even an internal PPT instructor at Tencent—so hasn’t my expertise already been replaced by AI? Where’s the value in any of it now?

But I think it’s exactly those years of front-line experience that let me direct AI this efficiently through a proposal design this complex. My understanding and experience here are already around ninety-five out of a hundred, and that’s why AI, as an amplifier, could produce an exponential leap.

I keep seeing the same thing: the industry veterans around me get a step-change in productivity out of AI, as long as they can turn their experience into thinking and requirements AI can follow.

Today I’m just using AI decks as an example. Plenty of AI applications have fallen into the same trap: they walk the old workflow step by step and throw engineering at each step to replace it. The question they should be asking is what sits behind the delivery, behind the action—what’s the real need?

All we ever wanted was a point that has been powered, so it can move our audience and move our clients. We were never after an editable PPTX file.

Source: EP0061_audio.mp3 · ASR model gemini-2.5-pro (chunked parallel) · full text of the original recording

[00:00] I think today’s AI deck tools—every single one of them—are pointed the wrong way. Why do I say that? In my seventeen, eighteen years of working, a huge share of my time went into planning and strategy, and the main job was writing decks. Whether it was a routine pitch or the slides for a launch event, I averaged one or two hundred decks a year, not counting revisions. And what I see is that today’s AI deck tools have all wandered into one enormous blind alley: they’ve poured a ton of time and resources into figuring out how to generate an editable PPT.

[00:25] Some AI tools even market it as their big selling point, push it as the thing that sets them apart. I think it’s a total waste of time. Think about it: why do we have a file format like PPT in the first place? PPT is short for PowerPoint—it’s meant to power the point. When we’re speaking, we need something visual to underline our core ideas so the audience understands us better. That’s all. Everything else exists to make that one thing happen more efficiently.

[00:51] Which means: as long as it can take the points I want to make in my talk and lay them out visually, page by page, fast, and then when I need changes, make them fast and the way I want them—that’s enough. We don’t care whether you can drag things around with a mouse, whether you can edit individual words. That doesn’t matter. And truthfully, I haven’t opened a piece of software like PPT in about a year, because all of my external teaching and all of my pitches are done as web pages.

[01:16] You can present them as slides on the web perfectly well, it supports a clicker, and with just a link I can fire it off to everyone in the room fast and they get a complete, clean version. So my process for making a deck with AI now looks roughly like this. First I have Claude Code ask me all kinds of questions, and I give it whatever I want to cover in this pitch or this class—don’t think about it, just throw it at it—so it can fill in all the information the deck needs.

[01:41] Things like: who’s the audience? What does the audience want, what are their pain points? And what’s the goal I’m trying to hit? Once that first round is done, I have the AI go do multi-angle, multi-agent parallel research to fill this deck out with solid detail. Of course, along the way I’ll also tell it the core ideas I want this pitch to get across, and even the order I want to tell the story in—it’s not entirely AI’s call. But the process becomes very relaxed. It’s just a chat,

[02:06] and you’re done. At the same time I have the AI produce what we call the black-text-on-white “big-print” web version. Every page is clickable so you can see how the information on that page is built, and it also makes it easy to see how the whole talk breaks into major sections and where the thread runs. Then I can keep talking with the AI, revising the specific content inside those pages, and lock down a first text version. And in that process the AI’s revisions are extremely fast, because you know AI does web pages—it edits them in literally a minute.

[02:31] And it doesn’t just revise the specific words on each page the way I asked. It can also re-sequence the page structure in there because I changed my mind about the narrative—that’s all on the table. Once the black-on-white version is confirmed, once I’ve read it through and it fully matches what I want, I can have the AI use the GPT Image 2 API to generate, in parallel, a pure-image version of the deck. What does that mean? It takes the black-on-white version and turns it into something genuinely beautiful, with image elements,

[02:56] even charts, all in one consistent template style, and then it carries that same treatment across the whole thing. And in that process we can give it some template references—look, this deck is really well done—or hand it the client’s logo or brand elements and have it produce a treatment that matches the client’s color scheme. And even though every page that comes out is a single complete image, you can see the style consistency is very strong, and the page content

[03:21] strictly follows the black-on-white version we already signed off on. Do you know how long it used to take me to write a deck like this? Five to seven days at the absolute fastest to finish a beautifully laid-out, 50-plus-page course deck. And do you know how long the deck I’m showing you now took to generate? The night before, I was still working with the organizer on the outline in the black-on-white version. Then, on the way to the classroom in the morning, it took about fifteen minutes to generate the whole deck—and it came out this good. Sometimes it does get to me:

[03:47] all these years of honing a professional skill—I was even an internal PPT instructor at Tencent—and hasn’t that professional skill of mine already been replaced by AI? What was the value of everything I did over all those years? And I think it’s precisely because of all those years on the front line that I can steer AI through a project this complex, this fast. It’s exactly because my knowledge and experience here are already at, say, 95 out of 100 that AI, as an amplifier, could take it up exponentially.

[04:13] So more and more I find that the old hands around me, with AI behind them, as long as they can get their experience out in a form AI can understand—the thinking and the requirements—their productivity jumps a whole level. And really, today AI decks are just my example. A lot of AI products have fallen into this same trap: they take the old way of working and try to engineer a substitute for it, step for step. When the real question—the real question you should be asking is what the need behind that deliverable, that task, actually is. All we want is a point that’s been powered,

[04:39] so we can move our audience, move our clients. What we want is not an editable PPTX file. That’s it for today. Follow me, and see you next time, bye.