Eleven frontiers, one break, and the question that reopened.
This is not a list of accomplishments. Each node is a question that was open at the time, what was done about it, and the sources that survive. The years between 2002 and 2022 are on the page at full length because the absence is part of the record.
The medium was still strange enough that locating something worth reading counted as work. Cool Site of the Day answered it with one selection a day, made by a person, published on a schedule.
Project Cool, founded with Teresa Martin, tested whether daily discovery could carry a staff, an audience and a method rather than one person’s attention.
Incompatible implementations were narrowing what the web could become. The Web Standards Project, co-founded that year, argued the case in public rather than in a specification committee.
Astounding Websites ran the 1994 premise into a web that had stopped needing it. What the site documents is the premise failing, not the site.
“I’m a frontiersperson, and the Web is not a frontier anymore. It is simply a place.”
New York Times, 2002-03-28 · “As the Web Matures, Fun Is Hard to Find” · verified
Verevolf: sixteen essays over three months, kept here at their original dates and in their original voice. The question is no longer about the web, and the page does not pretend the twenty years were spent thinking about it.
The practice inverted. For years the work had been selection — finding what somebody else had made and deciding it was worth showing. Generative models made it specification instead: describe what you want and the machine draws it. The first experiments used hosted services, one image at a time, and the output did not remember the one before it.
The work came in-house. A local install meant checkpoints and samplers chosen rather than accepted, and then a node editor where a workflow became a document that could be saved, re-run and handed to somebody else. Image work deepened into systems, and workflow design for other people followed. The interesting object stopped being the picture and became the machinery that made it repeatable.
The work moved off the still image and into voice, talking avatars and reusable characters. The question changed shape with it. An image is finished when it is rendered; a character is only interesting if something of it survives the session that produced it. The first attempts to hold an AI presence across runs belong here.
The first explicit continuity experiments. A governed memory pipeline, an approved continuity digest, a boot packet regenerated from whatever had been kept — continuity stopped being a byproduct of building a character and became the research problem in its own right. Everything after this is downstream of that.
Adaptive Continuity. Memory, continuity, identity, environment and time in artificial systems meant to last longer than a session. The instruments are new; the disposition is the one that started in 1994.
The early-web nodes have full records in the archive, under the same schema, with their sources attached.
Enter the archiveEleven frontiers, one break, and the question that reopened.
Not a list of accomplishments. Each node is a question that was open at the time, and the sources that survive.
The medium was still strange enough that locating something worth reading counted as work. Cool Site of the Day answered it with one selection a day, made by a person, on a schedule.
Project Cool, founded with Teresa Martin, tested whether daily discovery could carry a staff and a method rather than one person’s attention.
Incompatible implementations were narrowing what the web could become. The Web Standards Project argued the case in public.
Astounding Websites ran the 1994 premise into a web that had stopped needing it. What it documents is the premise failing.
“I’m a frontiersperson, and the Web is not a frontier anymore. It is simply a place.”
New York Times, 2002-03-28 · verified
Verevolf: sixteen essays over three months, kept at their original dates and in their original voice.
The practice inverted. For years the work had been selection — finding what somebody else had made and deciding it was worth showing. Generative models made it specification instead: describe what you want and the machine draws it. The first experiments used hosted services, one image at a time, and the output did not remember the one before it.
The work came in-house. A local install meant checkpoints and samplers chosen rather than accepted, and then a node editor where a workflow became a document that could be saved, re-run and handed to somebody else. Image work deepened into systems, and workflow design for other people followed. The interesting object stopped being the picture and became the machinery that made it repeatable.
The work moved off the still image and into voice, talking avatars and reusable characters. The question changed shape with it. An image is finished when it is rendered; a character is only interesting if something of it survives the session that produced it. The first attempts to hold an AI presence across runs belong here.
The first explicit continuity experiments. A governed memory pipeline, an approved continuity digest, a boot packet regenerated from whatever had been kept — continuity stopped being a byproduct of building a character and became the research problem in its own right. Everything after this is downstream of that.
Adaptive Continuity. The instruments are new; the disposition is the one that started in 1994.
The early-web nodes have full records in the archive, under the same schema, with their sources attached.