flowchart TB
subgraph WRITE["Compressed at write time: once, lossily, for everyone"]
direction LR
A["Full analysis<br/>workings, dead ends, caveats"] -->|"author discards"| B["12 slides"]
B --> P1["Partner"]
B --> P2["Analyst"]
B --> P3["Sceptic"]
end
subgraph READ["Compressed at read time: per reader, reversibly"]
direction LR
C["Full analysis<br/>kept in the artifact"] --> D{"Reader picks<br/>the depth"}
D --> Q1["Partner<br/>the twelve pages"]
D --> Q2["Analyst<br/>opens the workings"]
D --> Q3["Sceptic<br/>opens the appendix"]
Q2 -.-> C
Q3 -.-> C
end
classDef lost fill:#f8f9fa,stroke:#95a5a6,color:#2c3e50
classDef live fill:#eafaf5,stroke:#18bc9c,color:#2c3e50
class A,B,P1,P2,P3 lost
class C,D,Q1,Q2,Q3 live
style WRITE fill:#ffffff,stroke:#dfe4e6,color:#7f8c8d
style READ fill:#ffffff,stroke:#18bc9c,color:#16a085
1 Everyone Important Told You to Stop. You Didn’t.
In 2003, a federal accident board sat down with the slide deck NASA engineers had used to assess whether a chunk of foam had fatally damaged Columbia’s wing. The board’s verdict on the format was blunt: PowerPoint was unsuitable for engineering reporting, and NASA should go back to writing technical reports (CAIB Report, Vol. 1).
Edward Tufte took the same slide apart line by line: the low resolution, the disproportionate authority handed to a slide title, the nested bullets flattening a life-or-death finding into something you could skim straight past (PowerPoint Does Rocket Science).
The year after the accident board reported, Amazon banned decks outright. Bezos set out the reasoning at length some years later: slides “give permission to gloss over ideas, flatten out any sense of relative importance, and ignore the interconnectedness of ideas.” Senior team meetings open with everyone reading a six-page memo in silence before anyone speaks (CNBC, 2018).
So: an accident board investigating seven deaths, one of the most respected information designers alive, and a company on its way to becoming one of the most valuable in the world. All aimed at the same target. All landing cleanly.
Now, what did you build your last client deliverable in?
If the best arguments in the world couldn’t kill the deck, the deck was never being held up by arguments.
This isn’t a funeral for slides. It’s an argument about how they get made, and who gets to make them.
2 A Slide Is a Compression Artifact
Something else was holding it up.
In 1990, producing something bespoke, branded and interactive for a client cost more analyst-hours than any single insight was worth. Not slightly more. Orders of magnitude more. So you compressed. Twelve pages. One idea per page. One shape per idea. Everything else went into the appendix, or the bin, or that part of your brain that still quietly resents it.
Given the production economics of 1990, that was the only sane call.
Then the profession did what professions always do with a constraint they’ve lived under long enough. We promoted it. The ration became a discipline, the discipline became a craft, and eventually we were running training courses on the art of the single-slide message, as though squeezing an eight-week engagement through twelve rectangles were a communication principle rather than a budget one.
It was never a principle. It was a price.
The deck compressed once, at write time, lossily, for everyone.
The partner who needs the answer and the analyst who needs the workings receive precisely the same twelve pages, because producing two artifacts cost twice as much as producing one. So you built for the most senior person in the room and let everybody else make do.
Now hold that against a different question. What happens when compression can happen at read time, per reader, reversibly? The partner gets twelve pages. The analyst who wants to know why clicks through to the workings. The sceptic opens the appendix that no longer has to be an appendix.
Same artifact. Same effort. Not a better deck. A different kind of object.
3 Twenty Years of Escape Attempts (All of Them Failed)
None of this is a new idea, and that should worry you.
You could render a slide in a browser, free, in 2005. reveal.js has been around since 2011 and is genuinely good. R users had xaringan. Quarto ships revealjs today. The rest of the graveyard is well populated.
Business never showed up. Not “adopted slowly” — didn’t show up. Two decades of free, capable, open-source HTML presentation tooling, and the consulting deliverable stayed exactly where it was.
So here’s the comparison worth sitting with. On 28 January 2026, someone published a coding-agent skill that generates self-contained HTML presentations and converts PowerPoint files into them. As of 15 August 2026 it has 27,529 stars and 2,233 forks (frontend-slides). It is under seven months old.
Same output format. Same browser. The same kind of self-contained HTML file reveal.js has been producing since before some of your analysts could drive.
One caveat you should insist on: stars are interest, not adoption. The people starring a coding skill aren’t consultants, and twenty-seven thousand developers tell you nothing about whether your firm’s next QBR goes out as HTML. What the number shows is narrower and more interesting. The moment the capability stopped requiring somebody to learn it, demand appeared immediately. Whether that converts is a separate question, and there’s evidence further down.
The thing that changed wasn’t the format. It was that you stopped having to learn it.
There was a second cost, and it’s the one everybody misses. You could always make a slide render in a browser. What you could never do was make it your firm’s slide. The right navy. The right type scale. The logo where the brand team signed off on it.
Go and look at the branded reveal.js themes that exist in the wild. Hand-built. One per institution. Usually by a single enthusiast who happened to know Sass.
That isn’t a format. That’s a craft project.
And it explains PowerPoint better than any critique of it ever has. Nobody in a corporate designs a slide. They open an approved template and fill in the placeholders. The template was always a style engine. It just ran on people.
4 What Actually Changed (Two Barriers, Not One)
“AI makes it look nice now” is the lazy version of what happened, and it’s wrong in a way that matters. Two separate barriers fell, at different times, by different mechanisms.
The first was design skill — and the mechanism is more interesting than the model having good taste. You were never bad at taste. You were bad at describing it. So the tools stopped asking. Generate three visual options, let the user react, iterate. Showing beats specifying. Anyone who has tried to brief a designer over email already knew this.
Now the nuance that proves these really are two different barriers: that same skill ships presets and templates with names like “Editorial Forest” and “Neon Cyber”. They’re lovely. No brand team on earth has ever signed off on Neon Cyber.
Aesthetics fell first. Conformance needed a separate attack, and it got one.
_brand.yml is a single YAML file holding your logos, colour palette, semantic colours and typography. Quarto detects it and applies it across HTML, dashboards, revealjs slides, and Typst, which is to say a web page, an interactive dashboard, a slide deck, and print.
One brand declaration. Four render targets. Enumerated in the documentation.
That’s the entire argument arriving early. Posit also ships it as an agent skill, which closes the loop: conformance stops being a project you scope and becomes a step you delegate.
The agent never puts the logo four pixels off.
flowchart LR
IDEA["An analyst with<br/>something to say"] --> B1{"Barrier 1<br/>design skill"}
B1 -->|"fell: show options,<br/>stop asking for a spec"| B2{"Barrier 2<br/>brand conformance"}
B2 -->|"fell: declare the brand once,<br/>the renderer reads it"| OUT["Client-ready,<br/>on-brand artifact"]
OUT --> R1["Web page"]
OUT --> R2["Dashboard"]
OUT --> R3["Slides"]
OUT --> R4["Print"]
classDef gate fill:#ffffff,stroke:#95a5a6,color:#2c3e50
classDef good fill:#eafaf5,stroke:#18bc9c,stroke-width:2px,color:#2c3e50
classDef plain fill:#f8f9fa,stroke:#dfe4e6,color:#2c3e50
class B1,B2 gate
class OUT good
class IDEA,R1,R2,R3,R4 plain
Which brings the placeholder point back around. The corporate template was always a style contract, a machine-readable spec that happened to be executed by a human with a mouse. Nobody needed to invent the contract. We just needed something that could read it.
You’re inside one of these right now, incidentally. This page is a Quarto-rendered HTML artifact, and it required no design skill and no brand team.
5 The Flip
All of which adds up to a change in what derives from what.
Today: analysis → the deck is canonical → an HTML version exists if somebody’s feeling keen.
Next: analysis → the HTML artifact is canonical → the deck is one rendered view of it, and the workbook is an export button.
flowchart LR
subgraph TODAY["Today"]
direction TB
T1["Analysis"] --> T2["Deck<br/>canonical"]
T2 -.->|"if somebody's keen"| T3["HTML version"]
end
subgraph NEXT["Next"]
direction TB
N1["Analysis"] --> N2["HTML artifact<br/>canonical"]
N2 --> N3["Deck<br/>a rendered view"]
N2 --> N4["Workbook<br/>an export button"]
N2 --> N5["PDF / Canva"]
end
classDef canon fill:#eafaf5,stroke:#18bc9c,stroke-width:2px,color:#2c3e50
classDef plain fill:#f8f9fa,stroke:#95a5a6,color:#2c3e50
class T2,N2 canon
class T1,T3,N1,N3,N4,N5 plain
style TODAY fill:#ffffff,stroke:#dfe4e6,color:#7f8c8d
style NEXT fill:#ffffff,stroke:#18bc9c,color:#16a085
And this is where I stop predicting and start pointing, because it’s already shipping. Anthropic launched Claude Design in April 2026. Its own description of what you do with the result: share it as an internal URL, save it as a folder, “or export to Canva, PDF, PPTX, or standalone HTML files.” The frontend-slides skill goes the other way and converts PowerPoint into HTML, the flip with a direction arrow drawn on it.
This isn’t a forecast. It’s a changelog.
Read that export list again. The artifact is the thing you made; PowerPoint is one of the things you can turn it into, sitting in a list beside Canva and PDF, none of which anyone would mistake for where the work happened.
Meanwhile Prezi, which sells a PowerPoint alternative, publishes a 2026 buying guide scoring AI deck tools on “PPTX export fidelity”. Take the source for what it is; the framing still holds. The incumbent format has become a compatibility checkbox on a challenger’s scorecard, marked out of ten. Nobody rates Word on its ability to export to WordPerfect.
To be unambiguous: what’s dying is the way a deck gets made, not the deck.
The slide format is fine. Genuinely good, in fact: sequential, paced, one idea per view, survives a projector and a bad room. You will still present slides.
You just won’t open PowerPoint to build them. PowerPoint is demoted from an authoring application to an output format — from where you make the thing, to one of the shapes it can take.
6 The Handoff Was the Real Compression
There’s a second compression in this story, and it’s the one you actually feel.
Look at how analytics consulting is organised almost everywhere. The analyst takes the requirement, does the work, builds the visuals. Then hands off to the client-facing senior consultant, who assembles the deck, writes the storyline, supplies the interpretation, and carries it to the client.
That handoff is where the real compression happens. The analyst’s full understanding — every dead end, every caveat, every this is interesting but I couldn’t chase it — gets squeezed through a slide-shaped hole. Someone who didn’t build the charts then reconstructs a narrative from the residue.
We never called that lossy. We called it a process.
And it existed for a good reason, not a political one. The analyst genuinely couldn’t produce a client-ready narrative artifact: no design ability, no brand access, no client context, no time. So the work got divided at exactly the point where production cost peaked. Every consultancy arrived at the same split because every consultancy was handed the same bill.
The deck compressed the analysis. The handoff compressed the analyst.
Remove the production cost and that split has no particular reason to sit where it sits.
This isn’t a story about the senior consultant becoming unnecessary. That’s the lazy read, and it’s wrong. The storyline work doesn’t vanish; it relocates to whoever holds the domain knowledge and the client empathy. Sometimes the consultant, sometimes the analyst, increasingly whoever has both. What changes is that the format stops being the gate deciding who’s allowed to tell the story.
And be precise about what the machine does here, because the honest version is less thrilling than the pitch. AI is good at proof-reading, at drafting, and at generating insight from data a human has already curated. That word carries serious weight. The curation, the industry knowledge, the instinct for what this client will care about: that stayed human. It was always the hard part, and it still decides whether the artifact is any good.
7 Why Excel Is Bulletproof and PowerPoint Isn’t
So does the same thing happen to Excel?
No, and the reason isn’t habit or politics. It’s structural.
In Excel, the artifact is the workspace. The grid is input, computation and output simultaneously. When you send someone an .xlsx, you aren’t sending a finished document. You’re sending a working model they can poke and re-run with their own assumptions.
A deck is the opposite: a finished rendering, authored in one tool and consumed somewhere else entirely.
That separation is exactly what makes a deck’s authoring layer swappable. You can change how a rendered artifact gets made without changing what it is. You cannot do that to a live medium, because there the tool is the artifact.
A grid can only be replaced by another grid.
The obvious objection is Google Sheets, and it makes the point rather than dents it. Sheets moved the grid into the browser, which is precisely the migration this article says is happening to documents. It changed almost nothing about what a spreadsheet is. It didn’t demote Excel to an export format; it became a second grid, with an .xlsx door running both ways. Move a live medium to a new substrate and you get another instance of the same tool. The form survives the substrate. That is not what is happening to the deck.
Which is also why the “Excel is finished” genre has such a dismal record. In October 2025 the trade press reported a Gartner forecast that over 70% of finance organisations would be off spreadsheets as their primary planning tool by 2026.
The deadline hasn’t passed, and I won’t pretend otherwise: it’s August 2026 and nobody gets to call it yet. But the nearest survey to hand, from November 2024, still had 58% of finance leaders naming spreadsheets their primary tool. It predates the forecast, so it can’t falsify it. It just shows no sign of the run-up such a shift would need.
Not a cheap shot at Gartner. A note on what confident predictions about entrenched formats have to survive. Including mine.
There’s a better precedent still. Power BI arrived to replace the spreadsheet and was widely adopted. It then built, and kept expanding, an Export to Excel button, with Microsoft’s own release plan documenting round after round of improvements to it.
The challenger won the category and still had to ship a door back to the incumbent.
Excel didn’t survive by being better. It survived by being everywhere. It’s already installed, everyone has opened one, and the file lands in a format the next person can open too. Even the tool guides written for people entering the field still put SQL and Excel at the top of the list.
Ubiquity compounds. It doesn’t need a champion.
And when a format genuinely does die, it’s because the ground moved underneath it. Lotus 1-2-3 controlled 70 percent of the $500 million spreadsheet market in 1988. It didn’t lose to a better spreadsheet. It bet on OS/2 while Microsoft pivoted to Windows, and was left on a character-based interface when the world moved on.
Note which way that cuts. The substrate for reading business documents has already moved to the browser: an argument for HTML that requires nobody’s format to die.
8 What’s Actually Holding It Back (The Unglamorous Bit)
The honest obstacles aren’t conceptual. They’re plumbing.
Start with the one nobody mentions. A .pptx sails through your client’s email gateway. A self-contained .html file often doesn’t.
That’s not a misconfiguration, it’s deliberate, and it’s correct. HTML attachments are a well-known credential-phishing vector, and controlling which file types reach an inbox is baseline hardening: Australia’s Cyber Security Centre publishes attachment-filtering guidance for exactly this class of risk, and Microsoft’s mail protection detonates attachments in a sandbox before delivery. Whether a given gateway strips your file, holds it, or waves it through is a per-client lottery you don’t get to influence.
Which is the sharpest problem in the whole argument. The artifact’s best feature is the reason security blocks it. One file, portable, offline, opens in anything, still works in ten years — that is also, precisely, the shape of a phishing payload.
It gets slightly worse before it gets better: some tools’ “standalone HTML” export isn’t one file at all, it’s a folder of index.html plus assets. You can’t attach a folder either.
Which reframes the link-first approach these tools have taken. Not a fashion or a lock-in play, but a response to a constraint. The real distribution answer is a link, a portal, SharePoint, or an exported PDF. Not an emailed HTML file.
Though on longevity the challenger wins on the incumbent’s own turf. A .pptx needs PowerPoint; a single HTML file needs a browser. As the frontend-slides README puts it: “Dependencies are debt. A single HTML file will work in 10 years.”
Then the client mandate, which is institutional rather than technical. When someone must present to their board in the company’s approved template, they aren’t editing your artifact. They’re rebuilding inside a container they don’t control.
And the objection with real weight: Microsoft can do all of this inside PowerPoint. Copilot already builds a deck from a prompt, and Microsoft has documentation devoted to keeping the output on-brand. That’s both of my barriers, attacked from inside the incumbent, by the company holding the template estate, the identity system, and an install base no challenger can buy.
I take this seriously. It’s the most likely way I’m wrong.
But notice what it would and wouldn’t change. Copilot makes the existing model cheaper without making it different: one artifact, compressed once at write time, for everyone. No reader choosing their own depth. No single brand declaration rendering out to a web page, a dashboard, a deck and print. A faster horse, and a genuinely good one.
So this isn’t a caveat I’m parking. It’s the test, and it’s stated below with a date on it.
Finally, the objection you’ve been forming for several paragraphs: doesn’t the slide limit force discipline? Isn’t infinite scroll just beautifully rendered mush?
The 200-slide appendix deck exists. You have been buried by one. PowerPoint’s discipline is folklore.
9 What to Do Monday
Start with the delivered companion, not the flagship.
Not the working spreadsheet. Keep modelling in Excel; that’s the live medium and it isn’t the target. I mean the thing you send: the artifact a client opens to explore the findings. No template mandate, no board meeting, no brand sign-off, and interactivity is exactly what it’s for. It’s the easiest win on the table and almost nobody has taken it.
The flagship deck goes last, because what holds that format up was never the tooling.
Then generate for editability. The client has to be able to take it apart, or you haven’t shipped a deliverable. You’ve shipped a demo.
- Readable source. No minification. An executive with Notepad and Ctrl+F can change a sentence and reopen the file. That’s no scarier than editing a slide.
- Main narrative as plain text. Inline styling markup is fine; spans and classes are HTML doing its job. What you’re avoiding is the narrative living inside a JavaScript data structure, where nobody can find the sentence they want to change.
- Mark sections in plain comments.
<!-- SECTION 7 START -->. Cutting one should be as easy as deleting a slide. - Make charts exportable. Download as PNG or SVG. Better than PowerPoint, where the chart is usually a locked object welded to a data table you were never given.
- Assume snipping. People screenshot visuals out of decks and PDFs already. Make yours worth screenshotting.
And write the brand down once. A _brand.yml, a design-token file, a skill: the format matters far less than the principle. Conformance should be something you declare, not something you redo.
10 How I’d Know I Was Wrong
Three ways, all dated and checkable. But they don’t run on the same clock, and the gap between the two is the part worth arguing about.
The vendor clock — by end-2027. If the leading AI presentation tools still generate .pptx natively, rather than generating a web artifact and exporting .pptx from it, the direction of derivation never reversed. This is the Copilot question from two sections back, and it’s the cheapest test here: open the tools, look at what they emit. Nobody’s procurement committee gets a vote.
The buyer clock — by end-2029. Two conditions, both about firms rather than software:
- If the analysis companion shipped alongside consulting deliverables is still overwhelmingly an Excel workbook, with no real HTML analytic-report practice, then I was wrong about the case I called easy, and the hard case never mattered.
- If firms are still hand-building per-brand themes instead of declaring a brand once in a machine-readable spec, then conformance was never solved and the spine of this argument collapses.
The buyer clock gets the longer run deliberately, and not as insurance. In regulated sectors — healthcare, banking, government — the constraint on a new deliverable format was never whether the technology worked. It’s that the format has to clear the function whose job is to say no. Recall who blocks the HTML attachment, and that they are right to block it. The same mandate that correctly hardens the perimeter also sets the adoption speed, and adoption speed is not a function of how good the artifact is.
Which means the failure I’d actually expect isn’t the thesis breaking. It’s the tools flipping years before anyone’s template estate notices.
Gartner put a number and a date on a format prediction, and that date lands inside the next few months. Watching how it goes is exactly why I’m predicting a demotion, and not a funeral.