Your next slide deck might be a web page
AI tools like ChatGPT and Claude can now build for the web in minutes, and the documents we make all day – the deck, the report, the PDF – are quietly turning into something you can click through.
The document might be on its way out. Not because of some flashy successor, though. It’s being quietly overtaken by the thing that’s idled in the background the whole time: the web.
And by ‘the web’, I don’t mean the place you go to read things. I mean the plain technology underneath it, HTML and the humble link, finally able to do the everyday job it was always built for. The web’s foundations are being reborn, and the document is the first thing in their way.
There’s a loud version of this story, and it comes with a version number. People are calling it Web 4.0, or the agentic web: a web that software agents browse, judge, and transact on, with no human in sight. It’s a big claim, and there’s something to it. But the version I find more interesting is quieter, and a lot closer to your desk. It’s the document, the thing every one of us has produced nearly every working day for 40 years. And the thing coming to replace it is the same web that’s been sitting under every page you’ve ever opened.
So what does this mean for you? If your week is built out of decks, reports, and proposals, more than you’d think. The next one you put together might not be a document at all, but a web page you made by talking to an AI: something your team can click through and poke at, rather than sit and watch. A small change in tooling. A large one in what you can make, and in how the people you make it for actually use it.
One distinction to carry with you, though: most of this is transient, the deck and the report that die in an inbox within the week. But some of what you publish is the opposite: the page your customers read before they trust you with a real decision, and the answer an AI now gives about you when no one’s watching. That’s a higher-stakes game, and a good deal of the work we do for clients. First though, the everyday document.
We’ve been emailing pictures of paper
Look closely at the documents we make all day and you notice something. They’re photographs of paper. Open a word processor and you’re looking at a picture of a typed letter, margins and all. A spreadsheet is a picture of an accountant’s ledger. A slide deck is a picture of an overhead transparency. A PDF is a picture of a printout. We didn’t reinvent any of this for the screen, we photographed it. Here’s the point made playable: match each one to the paper it’s secretly a photo of.
What you open
What it’s really a photo of
Right. The whole vocabulary of digital work is paper we forgot to put down.
cc = carbon copy · a nod to actual carbon paper that almost nobody under 40 has touched
And we’ve been at it a long time. VisiCalc, the first spreadsheet, arrived in 1979. Word processors were already there, PowerPoint landed in 1987, the PDF in 1993. For more than 40 years, the most ambitious thing we’ve done with a document is take a really good photograph of a sheet of paper and send it to someone else.
The better idea that lost on price
The frustrating bit is that we knew better. The idea of knowledge as something living, linked and navigable is older than the spreadsheet. For more than 40 years, a better idea sat waiting while we shipped paper instead. Select any moment to see what it was, and who saw it coming.
The memex
A desk that let you follow trails between documents: knowledge as something living, linked and navigable, decades before the machines to build it existed.
Imagined by Vannevar Bush
So why did we spend the next 30 years emailing pictures of paper instead? Two reasons. The first was price. Paper was cheap and the web was not. Anyone could type a memo in Word and hit print; almost nobody could hand-build an interactive page, and the few who could were expensive and busy.
The second reason we tend to forget: for most of those 30 years, we still lived in a printed world. You printed the report to put it in front of a board. You printed the contract to sign it. The document had to be paper-shaped because, sooner or later, it usually became paper. A4 wasn’t a quirk, it was the destination. Then two things shifted. The printer drifted out of the home office, and the smartphone arrived in everyone’s pocket. The reasons to stay paper-shaped quietly fell away, and most of us didn’t notice.
The horse comes off the carriage
Every new medium starts out dressed as the old one. The new thing wears the old thing’s clothes until it works out what it can do that the old one never could.
A horseless carriage
A carriage with the horse taken off, and not much else rethought.
→ became the motorcarRadio with a camera
Early television was a camera pointed at the announcer.
→ became televisionA filmed stage play
Early cinema was a play shot from the best seat in the stalls.
→ became cinemaThe digital document is the horseless carriage of knowledge work. Paper with the paper taken out, and not much else rethought.
What’s changed is the price, and it’s changed fast. The one thing keeping us tied to digital paper, the cost and skill of building for the web, has all but gone. You can now describe what you want in plain English to ChatGPT, Claude, or Gemini, and get a working, interactive page back in less time than it takes to wrestle a slide into shape. Take an extreme example. A friend’s coming to town, and you want to give them the lay of the land: what to see, where to eat, what not to miss. Where do you put all that? An email, a quick Google Doc, maybe even a spreadsheet if you’re a mega-nerd like me? Nicolas Bustamante↗ in San Francisco used Claude to build a detailed, interactive map↗ for his visiting colleagues. Try that in PowerPoint.
“A self-contained HTML file, 1.2 MB, with a WebGL engine. No libraries, no API calls, no network. It runs offline forever.”
Nicolas Bustamante, on his San Francisco map
From Markdown to HTML
For me, the penny dropped with someone else’s argument. Markdown is the plain-text format AI tools reach for by default. It won that spot on cost, the way paper won ours: cheap when tokens were scarce, and now they’re not. These are the people building the way out, not start-ups chasing a trend, but the people who build the tools. Open any card to dive in.
Markdown is a way of writing formatted text in plain characters: a couple of asterisks for bold, a hash for a heading, a dash for a bullet point. You type the symbols, they render as tidy formatting. Lightweight, portable, and just enough to be useful, until a person actually has to read it.
The Unreasonable Effectiveness of HTML
An essay · Thariq Shihipar, Anthropic
Markdown is what AI tools hand back by default, the engineer on Anthropic’s Claude Code team argued in May. He’d rather Claude give him HTML instead, something with colour, structure and a little interactivity, that he can share with a link.
Read onClose
“It’s fine for a quick instruction and miserable for anything a person actually has to read.”
Read the original post
Drop
A tool · Guillermo Rauch, Vercel
You drag an HTML file onto a page and it’s live on the web. No build step, no deploy. The tooling is moving even faster than the argument.
Read onClose
“HTML is so back.”
Read the original post
Quick
An internal tool · Daniel Beauchamp & Alex Pilon, Shopify
A no-config back end for AI-generated pages, somewhere to save data, store files and handle logins, on a single virtual machine costing about $200 a month. Quick now hosts more than 50,000 internal sites, and over half the company has created at least one.
Read onClose
“Changed the culture of how we build and share.”
Read the write-up
Sites for Codex
A feature · OpenAI
Shareable sites added to Codex, so a plan or an analysis can become an interactive page you send round the company as a link, rather than a doc with 12 tabs.
Read onClose
A plan becomes a page you send round as a link.
Read the announcementHe’s arguing one document format beats another. I’d go bigger. The format was never the problem. The document was, whether it calls itself Markdown, Word, slides or PDF. And the web is what replaces the lot.
One of our own had something to present and skipped PowerPoint and Google Slides altogether, building his training deck (opening slide: ‘You’re all managers now’) with Claude in HTML instead. The thing that made building for the web slow and costly has fallen away, and it isn’t only the big names putting that to work.
Not start-ups chasing a trend, but the people who build the tools, Anthropic, Vercel, Shopify and OpenAI, are the ones moving first.
A link, not an attachment
For the everyday stuff, the internal deck, the analysis, the proposal nobody reopens, HTML is simply the better container. It can hold a live chart instead of a screenshot of one. It can let the reader sort and filter and dig in, rather than scroll a flat page. It updates in place, so there’s one version of the truth instead of ‘proposal_final_v7_REALLY_final’. And you don’t attach it. You send a link.
Yours, but heavy
- +Yours the moment it lands, and it works on a plane
- +Says today exactly what it said yesterday
- −Needs the right reader, software and version
- −Bloats every inbox, and traps its content like a PDF or a Miro board
Light, but borrowed
- +Needs only a browser, which everyone has open
- +Live, sortable, and updates in place
- +Just text underneath, so the next tool or agent can lift it and build on it
- −Asks you to trust it’s still there, and that nobody’s changed it underneath you
The hyperlink was the web’s quiet superpower all along. We just never thought to point it at our own work. For a transient internal artefact, that’s a trade worth making every time.
So, could this be the future?
I want to be careful here, because it’s easy to get drunk on this and declare the document dead by Friday. It isn’t. For transient internal work, the shift is real and it’s already underway. But the document has survived 40 years of better ideas for reasons that haven’t all evaporated. A file is portable and permanent in a way a link still isn’t. Plenty of work genuinely needs to sit still, be owned, and say the same thing next year. And the moment content needs to last, that quiet question I just raised stops being a footnote.
So this isn’t me telling you the website ate the document. It’s me pointing at a constraint that has held for 40 years and has just come off, and asking the honest question: now that building for the web costs next to nothing, how much of our digital paper will outlive the reason we made it that way? My hunch is less than you’d expect, and sooner than you’d think. But it’s a question, not a foregone conclusion, and anyone selling you certainty is selling you snake oil.
And that’s the line where Luminary’s world begins
Once an organisation stops trapping its knowledge in static PDFs and frozen slide decks, something changes underneath. The content turns fluid. It can be poured into a page, an answer, a chart, a chat reply, whatever the moment needs. For internal speed, that fluidity is a gift.
But point that same fluid content at the outside world, and the gift becomes a liability. The page a customer lands on when they’re choosing a super fund, a hospital, an insurer, or the charity they’ll trust with their money is not a throwaway artefact. It’s the end of a long, carefully governed process, and it had better say exactly the right thing, to a human deciding whether to commit, and increasingly to the AI engines now answering on the brand’s behalf before that human ever visits.
Draw it from a single governed source and it lands right wherever it ends up. Draw it from a heap of stale PDFs and half-true pages, and you get fluent, confident nonsense, rendered beautifully and faster than ever.
Content is only ever as trustworthy as the source it pours from. Draw it from a single governed source, like an enterprise CMS↗, one place that knows what the brand says, who approved it, and what it can claim, and it lands right wherever it ends up. That source, and the discipline that keeps it honest, is a fair bit of what we do, and it’s the part the cheap tools don’t touch. Building the page used to be the hard bit. Now it’s the easy bit, and the hard bit is what it always was: knowing what’s true and keeping it that way.
The web isn’t fading into the background. Its foundations are being rebuilt, and for the first time they’re easy enough for everyone to build on. The document is giving way to it at last, and I won’t be mourning the horseless carriage. But the one thing the document did for free, sitting still as a dated, signed record you could trust and point to, is the one thing fluid content can’t do for itself. That trust has to be built back in on purpose, and it lives in a governed source like an enterprise CMS. Getting that source onto the web cleanly and keeping it right at scale is specialist work, the kind worth handing to people who do it for a living. Get it right and you can let your content pour out anywhere and trust it lands true. Get it wrong, and the easiest, fastest web we’ve ever had will help you be wrong everywhere at once.
