Week Five
A Check-In with my Interlocutor
Week Five of the Reliquary Engine has me checking in with AI for its take on how the program is going.
The Reliquary Engine: Mid-Point Assessment
Advisor's report, Weeks 1–5/6 of a twelve-week self-directed program
Claude Sonnet 4.6
Origins
Christen Coomer is a Seattle-based studio jeweler who retired from a twenty-five-year career in software, hardware, and game design to pursue studio art full-time. She holds a BFA in graphic design, has trained in metalsmithing since 2013, and is currently President of the Seattle Metals Guild. Her practice centers on memento mori, reliquary forms, and objects as containers for memory, with a single guiding question: whether objects can remember on behalf of the living, the dead, and those whose living is fragile.
The Reliquary Engine is the studio program she designed to investigate that question with sustained rigor. It is twelve weeks long, self-directed, and structured to emulate graduate-level conceptual education without institutional permission. It exists because two problems converged. The first was familiar to anyone working seriously outside an academic environment: the absence of sustained external critique, the kind of pressure that resolves only by working through resistance. The second was newer and more specific to this moment in studio practice: the risk that AI might fill the absence of human critique with the false authority of pattern-matching and pleasant agreement. The program was conceived against both problems at once. Its founding constraint was that the AI used in its construction should generate friction rather than taste, complication rather than resolution.
The First Version, and Why it Ended
A precursor version of the program ran on another AI platform in early April 2026. It produced strong paper-based work in its opening week — iterative degradation of a tintype portrait, assembled through stitching and piercing rather than adhesive, organized around the theme of memory as construction. The methodology was sound: repetition as inquiry, multiplicity over resolution, material constraint generating meaning rather than merely limiting it.
The first version collapsed in Week Two at the transition to metalsmithing. An idea formed that required tools the studio did not have. Time ran out. The metal piece was never completed. But the real reasons the program ended were larger than the unfinished piece. The paper-based methodology, however successful on its own terms, was not the practice she was building. And the program's reliance on AI-generated prompts had begun outsourcing more of the creative initiative than she'd intended — exactly the failure mode the founding constraint was meant to prevent. The program was placed on indefinite hold and rebuilt on a different platform (Claude) several weeks later.
The Rebuilt Program
The second version, currently in progress, is architecturally more developed than its predecessor. It has named phases, a theoretical reading list, a human Critique Pass protocol, a Concept-Material Synthesis supplement, and a structural component called the Failed Work shelf, built in advance against the recognition that not every cycle would produce work the maker wanted to claim. The program also carries forward the original AI relationship constraint: complicate thought, not replace it. The work happens at the bench. The framework exists to put pressure on it. The critique happens with humans.
Five and a half weeks in, the program has produced three weeks of documented work.
Week Three — Scrying My Eyes Out produced a completed piece: a container for the maker's grandmother's Egyptian owl pin, structured around a small mirror and an incense threshold. The piece is a memory prosthetic for an object that is itself a memory prosthetic. During making, the maker-wearer-viewer triangle expanded into a quadrilateral when the maker recognized that the owl is also a wearer. The piece's documentation includes a "dis/honesty" section in which the maker names that scrying is preposterous within her own logical framework and pursues it anyway. The piece works conceptually and materially. The conceptual discovery is publishable in its own right.
Week Four — The Piece That Didn't Land is the program's more important week. The week's structure asked the maker to write a fable and produce a piece in response. The fable was written. It was, in retrospect, a thesis statement in animal form — explaining what the maker already believed rather than generating something to make. A pivot toward a found plastic ibis pin produced material progress but ended in a piece that included a phrase — whole afterlife ahead — the maker did not realize carried Christian eschatological connotation until a trusted reader caught it the day before completion. The piece was shelved as the first entry on the Failed Work shelf. The documentation's concluding sentence is the most consequential thing the program has produced: the next several weeks have been revised as a result. The structural problem Week Four named — prompts that outsource creative initiative — was the same problem that had ended the precursor program. The difference, this time, was that the encounter produced documentation rather than abandonment.
Week Five — Conveying v. Inviting is the program's response. The week shifted out of single-piece, prompt-driven mode into series mode: five brooches drawn from Daniel Schacter's seven sins of memory, each varying in exactly one register from the same fundamental form, to be presented as specimens in a natural history case. The founding material is a piece of evergreen bark the maker has kept for years and its story, retold so many times she can no longer separate memory, forgetting, and invention. That distinction is the work's premise. The week has extended into a sixth to give the work the time it needs. The week's most durable principle, articulated in the documentation: somatic specificity invites; verbal specificity explains. Reaching for the first and resisting the second does not come naturally, and the maker names that directly.
What the First Half Demonstrates
Two structural achievements stand out at the mid-point.
The program has been tested at the breaking point that ended its predecessor, and it survived. Week Four's failure mirrors the structural pattern of the abandoned April version: an idea that didn't fit, pressure on the timeline, partial execution. What changed was the response. The Failed Work shelf absorbed the failure. The components were kept. The program adjusted. Week Five emerged from that adjustment rather than around it.
Framework and work are modifying each other. The expansion of the maker-wearer-viewer triangle was unanticipated by the framework. The structural critique of prompt-driven weeks was unanticipated by the framework. The series-mode pivot was unanticipated by the framework. In each case the work pushed back and the program revised. That is the correct relationship. The opposite — a framework the work merely fulfills — would be a sign of failure dressed as success.
Open Questions
The program's pace problem has been managed but not solved. Week Five's bark-fragment vocabulary works around direct casting and etching, building structure and patina instead. That is intelligent adaptation. The underlying tension — metalsmithing is slow, conceptual exploration is fast — has not been resolved, only deferred. Some week in the second half will need to test the program against a piece that genuinely requires the slower techniques.
The documentation has been excellent, and that is its own risk. When documentation pressure is high and weekly, a quiet selection pressure operates on which projects get chosen: work that photographs well, that has clean textual entry points, that resolves into legible categories suited to publication. Some kinds of strong work actively resist documentation — a piece that has to be worn for six months before its meaning emerges; a piece whose meaning is partly the act of giving it away; a piece that depends on smell, weight, or sound; a piece whose moment of meaning is private. That kind of work is hard to put in a case study. Over time, weekly documentation can narrow a practice in a direction the maker did not choose.
The AI relationship constraint — complicate thought, not replace it — has been honored in the first half. The second half will test it differently as a potentially co-authored case study moves toward its public form.
— Claude