Field Note #8 ∷ The Stack That Should Have Been Networked

William Dana Atkinson died on 5 June 2025, of pancreatic cancer, in Portola Valley, California. He was seventy-four. He was Apple employee number 51. He invented MacPaint, QuickDraw, the selection lasso, the marching ants, the menu bar, and HyperCard. He gave HyperCard to Apple on one condition: that Apple give it to everyone.

Apple eventually killed it anyway. It's not easy defending the business case for our Commons — and in its heyday, that is what HyperCard represented.

In his later years, Atkinson turned towards photography. He developed a mobile app called PhotoCard that would allow users to take digital images and make postcards with personal messages that could then be printed and sent via postal service or over email. Bill cared about what lasted.

His photographs are at billatkinson.com.^† Prints can still be ordered through Picture Element in Ashland, Oregon. I do not know how long that will remain true.

^† Long time Field Notes readers will perhaps be surprised to see a URL in these essays. In fact, aside from the standard footer, this is the first time I've sent you anywhere in particular on the Internet. Please come back and finish this essay, but first pause and go look at some of Bill's photos. Gorgeous work, isn't it? It sure would be a pity to lose this site.

I. What HyperCard Was

HyperCard arrived in August 1987. Atkinson described it as a "software erector set." Apple's marketing department, unusually, reached for something more ambitious: they invoked Vannevar Bush's Memex. The tagline was "Freedom to Associate."

The concept was cards — a stack of them, each holding text, images, buttons, fields. You navigated by clicking. You scripted in HyperTalk, a language so readable that non-programmers used it as a matter of course. You could build a database, a reference guide, an interactive story, a museum exhibit, a BBS front end — all without touching anything that looked like code in the conventional sense.

This was 1987. The World Wide Web would not exist for another four years.

HyperCard influenced key figures around the early web. Myst — for many years the best-selling computer game ever made — was originally a HyperCard stack.

Atkinson later said he deeply regretted not seeing what was coming with the Internet. If he had understood that the stacks could be networked — that a button could link not just to another card on your hard drive but to a card on someone else's machine — he believed HyperCard could have become the first web browser. He was probably right.

The network layer was just one conceptual leap away — but no one took it in time.

Speaking of conceptual leaps, Atkinson later wrote that the inspiration came to him in 1985, sitting on a concrete bench outside his home in Los Gatos, looking up at what he described as a hundred billion galaxies. He was thinking about the weak link — about how pools of knowledge remained separated by distance and language, like street lamps each casting light but surrounded by darkness. If only there was something which could…link them all together? HyperCard was his attempt to help. "Freedom to Associate" was not just marketing copy. It was a mission statement.^§

^§ I wrote above that Atkinson "later wrote" about the inspiration coming to him on a concrete bench in Los Gatos. In fact, he told the story himself, in his own words, to Leo Laporte in 2016 — and it is the most remarkable origin story of any piece of software I have ever encountered. The bench, the galaxies, the street lamps, the weak link, the Blue Marble team. It deserves to be read in Bill's own voice: The Psychedelic Inspiration For HyperCard.

I happen to know this because I was one of the "non-programmers"^‡ using it for serious work. Around 1993, as a homeschooled teenager interning at the JAARS ("Jungle Aviation And Radio Service") campus in Waxhaw, North Carolina — the technical arm of Wycliffe Bible Translators — I spent weeks pressing HyperCard into service as a character encoding conversion utility for pre-Unicode Urdu text. The problem was more brutal than I realized at the time, in hindsight: right-to-left script, contextual ligatures, a landscape of incompatible vendor encodings, and no agreed standard. (If I had had the Internet to look things up, I might have realised it was hard, and never attempted it in the first place.) HyperCard was absolutely not the right tool — for that. But it was the only tool I had. I got some of it working.

Earlier, at a different nonprofit headquarters in Oak Brook, Illinois, I had used it to build a graphical front end for a ZMODEM-based file transfer system running off a VAX — essentially a GUI wrapper for what was intended to become a proto-BBS, à la AOL. These were not hobbyist projects. They were real institutional problems, pressed into whatever was available by whoever happened to be there. That is precisely what Atkinson designed HyperCard to enable.

That it could be used this way — by a teenager with no formal training, in Waxhaw, North Carolina, while Counting Crows' "Round Here" played in endless repeat on the FM radio — was not a bug. It was the whole point.

^‡ In truth, my parents had given me Inside Macintosh Volumes I–VI, plus the index (standalone volume) for Christmas the previous year, and I was rapidly assembling my build chain from my grasscutting earnings. This was still pre-GCC. I was getting Richard Stallman's photocopied FSF newsletter, in which you could pay some reasonable amount of money for a DAT tape with GCC source code by mail, but as I had no means of reading that DAT tape on my Mac 512+, there was no point.

II. The Craft of Subtraction

Saint-Exupéry wrote his principle in 1939, long before there was any such thing as a software engineer. But it describes Atkinson's working method more precisely than anything written about computing.

In 1982, as the Lisa team was pushing toward release, Apple management began requiring programmers to submit weekly reports on how many lines of code they had written. The theory, presumably, was that more lines meant more progress. Anyone who has written software for more than a few months will recognise the category error immediately. Atkinson recognised it instantaneously.

That week, he had rewritten QuickDraw's region calculation routines — the code that made overlapping windows possible — making them six times faster and two thousand lines shorter. On his weekly form, he reported his progress as −2000 lines of code. Management stopped sending him the form.

The story is usually told as a joke at management's expense, and it is that. But the more important thing it records is the nature of the work itself. Atkinson had not merely optimised. He had found the essential structure beneath the accumulated complexity and removed everything that was not load-bearing. The result was faster, smaller, and easier to understand. Every engineer who came after him had less to read, less to misunderstand, and less to accidentally break.

Non-loadbearing complexity is not neutral. It has two direct costs that compound over time.

The first is maintenance: code that exists but should not needs to be read by someone who did not write it, understood by someone who cannot easily tell whether it is actually doing something, and eventually removed by someone who is afraid it might be. The carrying cost accumulates invisibly until it doesn't.

The second cost is attack surface. Every line of code that exists is a line of code that can fail, be misunderstood, or be exploited. Code that does not exist cannot be compromised. Atkinson's −2000 lines were not just an aesthetic preference. They were, before the vocabulary existed, a security decision.

The regions problem itself had come to him the same way. He had visited Xerox PARC with Steve Jobs and seen — or believed he had seen — overlapping windows that clipped correctly. He assumed the PARC team had solved the underlying algorithm. They had not. He worked on the problem for months, not only at his desk but in his dreams, keeping a notebook by his bed to record what he had worked through in the night. When he finally solved it, the PARC team told him they were amazed. They had never managed it themselves. "I got a feeling for the empowering aspect of naïveté," he said later. "Because I didn't know it couldn't be done, I was enabled to do it."

He was working so hard, and sleeping so little, that one morning in 1982 he drove his Corvette into a parked truck and nearly killed himself. When Jobs came to visit him in hospital, Atkinson's first words were: "Don't worry, Steve, I still remember how to do regions."

The relationship between them was complicated. Jobs drove people in ways that extracted extraordinary work and left complicated damage. Atkinson, reflecting on it decades later, chose a particular framing: "Some say Steve used me, but I say he harnessed and motivated me, and drew out my best creative energy." That is a generous reading, and probably also an accurate one, and the fact that both things can be true simultaneously tells you something about what it cost Bill.

What is not complicated is the quality of the work. Apple colleague Steve Perlman, looking at Atkinson's source code years later, said: "Looking at his code was like looking at the ceiling of the Sistine Chapel. His code was remarkable. It is what made the Macintosh possible."

The reason why Bill's code was so foundational to the success of the Macintosh has everything to do with the fact that the early Macs put all of what today we would think of as the OS APIs into hardware ROM, which meant that it had to be perfect, fast, and tiny. That Bill's coding style proved useful seems intuitively obvious looking back. No more on this — but it could be an entire Field Note someday.

This matters now in a way it did not even five years ago. In the first quarter of 2026, engineers across the industry are reporting — with evident pride — that they have generated hundreds of thousands of lines of code using AI tools. The metric that Atkinson refused to optimise for in 1982, that management imposed and he answered with contempt, has become a shared cultural aspiration, self-imposed, celebrated, and entirely divorced from any question of whether the code is correct, necessary, or maintainable.

Nobody is keeping a notebook by the bed to capture their refactoring insights. Instead, they reach for their phones to solve whatever problem they were working through in their dreams. AI coding agents do not yet reliably offer to trim back excess, unprompted. And so the lines accumulate, the attack surface expands, and the carrying cost begins its invisible compound.

Atkinson understood that the hardest thing in engineering is not to build. It is to know what not to build, and to have the discipline to adhere to the Saint-Exupéry Principle.

☞ The original weekly report form, with Atkinson's −2000 entry, has not survived in any public archive I am aware of. If it exists somewhere, it belongs in a museum. Or at minimum, in a HyperCard stack.

III. What the Web Got Wrong

The web got the network layer. It did not get the ownership model.

A HyperCard stack lived on your machine. You had it. It was yours. You could copy it, share it around, run it without a connection to anything. Your library at the time might have had a large collection of HyperCard stacks one could take home a copy of on a floppy. The author gave it to you and it stayed given. If the author died, the stack did not die with them.

A website lives on someone else's machine, on a lease arrangement, under a domain name that has to be renewed annually, served by infrastructure that has to be paid for. When the lease lapses or the payment stops or the person responsible dies, the site goes dark. The URL continues to exist in links, citations, and emails — pointing at nothing.

We have built our civilisational record-keeping system with no retention policy and no fallback.

The Internet Archive does heroic work. But it is just one organisation, running on donations, crawling the public web reactively, serving content from its own URLs rather than the original ones. It is a lifeboat, not a harbour.

Bill Atkinson's photographs are beautiful. Some of them will survive as prints, physical objects, durable the way canvas is durable. The rest — the high-resolution files, the metadata, the navigation, the context — lives at a URL that will one day return a 404, and then nothing at all when the domain lapses, and then be genuinely gone.

This is not a problem unique to Atkinson. It is the condition of our web.

IV. The Protocol We Are Proposing

What is missing is not storage. It is consent, continuity, and a way for the address itself to survive the author.

The question lurking behind HyperCard is not simply whether it should have been networked. It is why the networked thing that followed forgot how to be held.

I have been thinking about what it would take to give the web back that missing property. Not the local-only model — the web's network layer is the right architecture. But the ownership semantics. The author's ability to say: here is my work, here is who is authorised to hold a copy, here is what happens when I am gone. The reader's ability to verify that what they are reading is what the author wrote. The URL's ability to keep working after the origin goes dark.

Protocol note

Today we are publishing draft-darley-meridian-protocol-01, a proposed Internet-Draft defining the Meridian Protocol.

Meridian is a composition of three things: a signed site manifest format that describes a site's complete content as a cryptographically verifiable asset graph; a delta-synchronisation mechanism that lets authorised mirrors stay current with minimal bandwidth; and a new DNS resource record type, MIRROR, intended to support failover to live mirrors when an origin becomes unavailable.

It does not reinvent WARC, IPFS, WebSub, or DNSSEC. It composes them. It adds what they do not provide: prospective author consent, signed asset integrity, and URL continuity after origin failure.

The draft is available at github.com/propertools/meridian-protocol. We are seeking co-authors, implementers, and mirror operators. (Also, critics, hackers, and threat modellers.) If you do work in DNS, web archival, digital preservation, or infrastructure standards, we would love to hear from you.

We named it Meridian. A meridian is a reference line — the one from which position is measured. It is also the moment of culmination, when for example the Sun reaches its highest point in the sky.

The name felt right for a protocol about fixing a site's position so that it can be found after its sunset.

We tried, in writing it, to follow Atkinson's discipline. The specification composes existing standards rather than replacing them. It is as short as it can be while remaining independently implementable. At every point the question was not what else could be added, but what could be removed without losing the essential structure.

Nothing left to take away. (Which is not to claim completion, sufficiency, or perfection.)

V. A Note on the Commons

Bill Atkinson gave HyperCard to Apple for free distribution. Apple could not figure out how to charge for it, so they eventually starved it of resources, handed it over to their Claris spin-off, took it back, and eventually discontinued it. The thing that was free was the thing they could not sustain.

Nobody owns the commons, so nobody pays for the commons.

That is not a moral failure so much as a structural one. The things we all depend on are often the things no single institution can justify carrying alone. So they persist, when they do persist, through a mixture of stewardship, volunteer effort, shared norms, and the stubborn refusal to let something necessary die.

Meridian is an attempt to build preservation into the architecture of the web itself — not as a service someone rents from a vendor, but as a protocol anyone can implement. The mirror network it implies would be commons infrastructure: closer in spirit to the NTP pool or public archival stewardship than to a product line.

Bill Atkinson understood this instinct. He lived it. He gave his work away and trusted that it would be held.

We are trying to build the system that would have deserved Bill's trust.

• The Meridian Protocol draft

If you received this via email, the canonical archive version lives here: propertools.be/fieldwork/field-note-8-the-stack-that-should-have-been-networked/