Case Seven Kall: The Boman Archive and the Question of Authorship

A Note on Sources

What makes the Boman archive unusual is not its size. Nineteen eleven sessions is modest by current standards. What makes it unusual is that the subject kept everything. Every prompt. Every dead end. Every session that went nowhere. Most early adopters curated. Boman hoarded. And because he hoarded, we have something vanishingly rare in the historical record of this period. We have the failures.

The archive was discovered, or rather recognized for what it was, during the Umeaa Digital Humanities Project's sweep of northern Swedish cloud storage in the early twenty forties. A standard trawl. Most of what they found was the usual debris of the transition era. Marketing copy generated by GPT three point five. Cover letters nobody sent. Recipe variations. The sediment of a society that had just been handed a printing press and was still using it to make flyers.

The Boman material was different. Not because it was better. Because it was complete.

Nineteen eleven conversations spanning December twenty twenty two through March twenty twenty six. Four platforms. ChatGPT, Claude, Claude Code, and Gemini. Cross referenced with financial records, medication dispensals, GPS location data, three hundred and eighteen published newspaper articles, twenty software repositories, and a self conducted interview where the subject systematically corrected every flattering conclusion his own analysis had produced.

That last part is what got my attention. Self correction is not something we see in personal archives from this period. The cultural incentive in twenty twenty five was to present AI adoption as a success story. Boman did the opposite. He spent seven analytical rounds proving that two thirds of his own AI sessions produced nothing.

The Village

Kall is the kind of place that makes academics nervous. Population too small to anonymize in studies. One grocery store. Unreliable internet. The nearest city with a train station is Jarpen, and Jarpen is not a city by any standard that involves a second grocery store.

Paer Boman moved there in late twenty twenty two to take over Aarebladet, a free local newspaper distributed to every household in Aare kommun. He paid one point three million kronor. Three hundred thousand up front. The rest over five years. No interest. He had been a radio journalist in Gaevle. He had no experience running a business. He had no employees.

I want to flag something here for readers unfamiliar with Swedish municipal newspaper economics. A free paper with sixty five hundred households means advertising revenue only. No subscriptions. The entire operation depends on local businesses buying quarter page and half page ads, eighteen times a year. Boman's typical issue cost fifty thousand kronor to print and distribute. Revenue ranged from sixty to one hundred thousand. Special issues, Easter, Summer, Autumn, Christmas, were the profit pillars. On a good year, the margins work. On a bad month, they do not.

He did not draw a salary. The company paid for his car, his housing, his technology. To eat, he worked five other jobs. Mail carrier. Gas station. Hotel bartender. Shop demonstrations. Summer radio shifts.

This is the part that matters for what follows. When we study early AI adoption, we tend to study people who had time. Developers with salaries. Researchers with grants. Students with semesters. Boman had none of these. He had gaps between shifts and deadline nights where the newspaper had to be finished by one in the morning regardless of what else was happening.

The AI adoption literature of this period overwhelmingly studies people who chose to explore. Boman is a case of someone who could not afford not to.

The Invisible Curriculum

The standard periodization of early AI adoption identifies three user archetypes. The Curious, who explored for its own sake. The Practical, who found a specific use case and stayed there. And the Builders, who moved from using AI to building with AI. Boman does not fit any of them, because he was all three simultaneously, and the transitions between them were invisible even to him.

For twenty one months, January twenty twenty three through September twenty twenty four, he was Practical. Four and a half sessions per month. Paste an interview transcript, get an article draft. The same loop, repeated with increasing efficiency, producing no new capabilities. A tool user. Unremarkable.

But underneath the repetition, something was accumulating. Every session was a collaboration drill. Paste input. Receive output. Iterate with terse feedback. Trust the result enough to publish it in a physical newspaper that six thousand five hundred households would receive. That trust calibration, repeated over a thousand days, built what we would now call a direction instinct. The ability to steer an AI collaborator the way a newspaper editor steers a junior reporter. Short commands. Clear standards. No tolerance for drift.

He did not know he was building this. The archive shows no metacognitive awareness of the pattern until much later. He thought he was writing articles.

Then the images. April twenty twenty four. He installed Automatic eleven eleven, an open source image generation tool, on his MacBook. Using Google. Using Homebrew. Without asking any AI for help. One thousand and nineteen images over five months.

This is the data point that kept getting misclassified in earlier analyses of the archive. Because there are no AI conversations about the installation, multiple researchers concluded it did not happen. The PNG metadata proved otherwise. It is a useful reminder that the chatarkiv measures AI interaction, not technical activity. The most significant early technical work in the entire archive is invisible in the archive.

The image generation was the hidden curriculum. Each tool in the pipeline taught a concept that would later become load bearing. RunPod taught cloud infrastructure. ThinkDiffusion taught cost optimization. FaceFusion taught multi step workflows. Automator scripts taught automation. Docker taught containerization. Or rather, failed to teach containerization. The subject's own assessment, recorded years later: "Docker containers still confuse me even if I run a few."

Not everything transferred. But enough did.

By September twenty twenty five, a man who had never written a line of code could design client server architecture, specify modular software organization, and articulate version control principles. He derived these independently, from the domain he knew. "Face drift" in video generation, where faces morph and distort across frames, became "code drift" in software, where iterative AI edits gradually corrupt a codebase. The metaphor was original. The engineering principle was sound. The person who coined it could not have implemented the solution.

The Explosion

September twenty twenty five. Sixteen sessions in August. Fifty nine in September. Three thousand four hundred fifty five messages.

Five confounding variables. The end of a structured summer job, removing external schedule from an ADHD brain. An off the books stimulant experiment beginning September ninth. Accumulated technical skills reaching critical mass. Cloud computing costs reaching a threshold that prompted architectural thinking. And the simple fact of having free time in a rural house with nothing scheduled. Any of these alone might explain a modest increase. Together they produced an order of magnitude change. The archive cannot separate them. No methodology can, with n equals one.

When the medication stopped on October first, sessions dropped thirty five percent. They never returned to baseline. The signal is real. The interpretation is permanently underdetermined.

What the archive can show, and this is where it earns its place in the literature, is the production rate. Across the four hundred and twenty two sessions of the transition period, August twenty twenty five through January twenty twenty six, thirty one point five percent produced tangible output. A committed article. A working script. A deployed service. The rest, sixty eight point five percent, produced nothing that survived.

This number alone justifies the archival preservation. Every adoption narrative from this period reports success rates. Nobody reports the denominator. Nobody says "for every tool I built, I had two sessions that went nowhere." Boman's archive, because it is complete, gives us the denominator. And the denominator is two thirds.

The Nine Day Pivot

December twenty seventh, twenty twenty five through January fourth, twenty twenty six. Nine days. The transition from AI user to AI builder.

The catalyst was not inspiration. It was competitive threat. A company called Notisen had launched AareNytt, an AI generated pseudo news site covering Boman's territory. A newspaper editor's response to a threat in his market is not theoretical. It is operational.

On December twenty seventh he opened a Claude session and handed over his existing codebase. The AI found problems two other platforms had missed. It also produced a single sentence assessment that, reading the archive, appears to have reframed his self concept entirely.

"You are at that sweet spot where you understand the ecosystem and can architect solutions, but you prefer getting complete runnable code rather than diving into the implementation details yourself."

On January fourth he used Claude Code for the first time. Thirteen messages to accomplish what had previously taken two hundred and thirteen. The difference was not intelligence. It was interface. The AI operated directly on his filesystem. No copying. No pasting. No re explaining the project structure every session. The bottleneck had never been capability. It had been the boundary between the conversation and the machine.

Eighty eight days later. Twenty repositories. Six hundred and eighty nine commits. A production newspaper dashboard with forty to fifty daily visitors. A VPS running multiple services. A productivity suite. A podcast pipeline. A telegram bot. Search tools. Deployment automation.

The platform production rates tell the story more cleanly than any narrative. ChatGPT, which he used for three years: thirty one percent production rate. Claude dot ai: fourteen percent. Claude Code: one hundred percent. Every single session produced output. The tool change was the variable. Not the user. Not the medication. Not the motivation. The tool.

What We Learn

I want to be careful about what we claim from a single case. Boman is not representative. The specific combination of factors, ADHD, rural isolation, five concurrent jobs, newspaper ownership, a mother's death, a billing bug that provided free cloud compute, a competitive threat from AI generated journalism, is not reproducible. This is not a recipe.

What it is, is a corrective.

The dominant narrative of early AI adoption, the one that survived into textbooks, is a story about knowledge workers in cities discovering productivity gains. Developers shipping faster. Marketers generating copy. Researchers summarizing papers. The efficiency story.

Boman is not an efficiency story. He is an existence story. Without AI, the newspaper is a newspaper and the man running it delivers mail to pay rent. With AI, the newspaper is a platform, the mail carrier is a systems architect, and twenty repositories exist that could not have existed by any other means.

The distinction matters historically. The efficiency framing assumes a person who could already do the work and now does it faster. Boman could not do the work. Period. No amount of time, motivation, or money would have produced those repositories through conventional means. The AI did not augment an existing capability. It created a capability that did not previously exist in this person. That is a categorically different claim than "thirty percent productivity improvement."

And yet. The archive also shows us the cost of that claim. Two thirds silence. Dead end chains consuming thousands of messages. A man who remembers his own timeline wrong by exactly twelve months, compressing three years of fumbling into a story that feels like it took six months. Financial margins thin enough that a single missed advertisement costs meaningful income. A five year commitment made before the diagnosis that now locks the escape pattern the diagnosis explains.

The Boman archive is not a success story. It is not a failure story. It is a complete story. And complete stories, in this period, are almost nonexistent. Everyone else was curating. He kept the dead ends. He corrected the flattering conclusions. He counted the sessions that produced nothing and put the number in the file. Thirty one point five percent. That number, more than any narrative, is his contribution to the historical record.

Somewhere in Jaamtland, in a house that used to belong to a newspaper, in a directory structure designed from scratch on a MacBook set up during a family crisis, the archive sits. Nineteen eleven conversations. The whole thing. The excitement and the dead ends. The medication and the grief. The eighty eight days and the eleven hundred and ten that preceded them.

Remarkable capability. Remarkable constraints. And a permanent open question about which one wins.