Doc to web page

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.

Adam Griffith

21 June 2026

12 minute read

View this blog post as an interactive web page (created by Claude).

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, that gridded green paper ruled into columns. A slide deck is a picture of the overhead transparency and the 35mm carousel it replaced. A PDF is the purest of the lot, a picture of a printout, frozen so it lands the same on every machine. We didn't reinvent any of this for the screen. We photographed it.

The language gives the game away. We still 'file' things in 'folders' on a 'desktop'. We 'cc' people, short for carbon copy, a nod to actual carbon paper that almost nobody under 40 has touched. The whole vocabulary of digital work is paper we forgot to put down.

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. Vannevar Bush described it in 1945, a desk that let you follow trails between documents. Ted Nelson named it 'hypertext' in the 1960s. In 1968, Doug Engelbart sat in front of a stunned audience and demonstrated the whole thing working, a mouse, linked documents, live collaboration on a screen, decades before any of it was meant to exist. When Tim Berners-Lee proposed the web in 1989 and built it over the next two years, he wasn't inventing something new. He was finally shipping an idea that had been waiting four decades for its moment.

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 motorcar arrived as a 'horseless carriage', a carriage with the horse taken off and not much else rethought. Early television was radio with a camera pointed at the announcer. Early cinema was a stage play filmed from the best seat in the stalls. The new thing wears the old thing's clothes until it works out what it can do that the old one never could.

The 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. And it's all just, in his words, a "self-contained HTML file, 1.2 MB, with a WebGL engine. No libraries, no API calls, no network. It runs offline forever."

Interactive map of San Fransisco

From Markdown to HTML

Quick aside for anyone who doesn't live in this stuff. 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.

For me, the penny dropped with someone else's argument. In May, Thariq Shihipar, an engineer on Anthropic's Claude Code team, published a piece titled 'The Unreasonable Effectiveness of HTML'. Markdown, he pointed out, 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. It's fine for a quick instruction and miserable for anything a person actually has to read: a plan, a code review, a report nobody opens. He'd rather Claude give him HTML instead, something with colour and structure and a little interactivity, which he can share with a link. 

Tiles with eight types of information: Tables, Design, Illustrations, Code, Interaction, Workflows, Spatial and Images. 
Two panels with one depicting markdown text an the other depicting the HTML output.

He's arguing one document format beats another. I'd go bigger. The format was never the problem. The document was the problem, whether in the form of Markdown, Word, slides or PDF. And the web is what replaces the lot.

The tooling is moving even faster than the argument. Vercel has just shipped a thing called Drop: you drag an HTML file onto a page and it's live on the web, no build step, no deploy. His reaction, in full: "HTML is so back." 

At Shopify, Daniel Beauchamp and Alex Pilon describe an internal tool their team calls Quick, which hands those AI-generated HTML pages a no-config back end (somewhere to save data, store files, handle logins) running on a single virtual machine that costs about $200 a month. Quick now hosts more than 50,000 internal sites, and over half the company has created at least one. Their verdict: it has "changed the culture of how we build and share". 

And OpenAI has added shareable 'sites' 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.

Codex dialog box

OpenAI’s Codex dialog box showing a prompt to create a dashboard app with its new Sites feature. 

Look at who's doing it. Not start-ups chasing a trend, but the people who build the tools: Anthropic, Vercel, Shopify, OpenAI. And it isn't only the big names. Closer to home, one of our own, Jase Watson, recently had something to present and skipped PowerPoint and Google Slides altogether, building his training deck with Claude in HTML instead. The thing that made building for the web slow and costly has fallen away.

A presentation slide with the heading 'You're all managers now'.

A slide from Jase’s presentation. 

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.

That difference matters more than it first appears. An attachment needs the right reader on the other end, the right software, the right version, and it still bloats every inbox and trips every spam filter. A link needs a browser, which everyone already has open. The hyperlink was the web's quiet superpower all along. We just never thought to point it at our own work.

View this blog post as an interactive web page (created by Claude).

And you're not the only one reading it now. A PDF or a Miro board traps its content: what goes in rarely comes back out. An HTML page is just text underneath, so the next tool or an agent can lift it, convert it and build the next thing from it, instead of starting from a blank page.

A link isn't a free lunch, mind you. A file, for all its faults, is yours the moment it lands. It sits in your inbox, it works on a plane, and it says today exactly what it said yesterday. A link asks you to trust that the page is still there, that you still have access, and that nobody's quietly changed it underneath you. 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.

Where it stops being simple

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.

Content is only ever as trustworthy as the source it pours from. Draw it from a single governed source (e.g. 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. Draw it from a heap of stale PDFs and half-true pages and you get fluent, confident nonsense, rendered beautifully and faster than ever. 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.

Want to tap into the expertise of an agency that’s been in operation since 1999?

Get in touch

Keep Reading

Want more? Here are some other blog posts you might be interested in.