Resident Inventor / companion program
How I Collaborate
With AI
This is how I deliberately add computational machinery to that process without handing over the part that makes it mine.
Playable model · v0.2
AI Collaboration
Program Builder
Build a short experiment or a long loop. The code is conceptual, not sacred. The vote stays with the person who knows what the work is trying to mean.
Your program
0 stepsBuild any sequence. Suggestions are guidance, not rules.
Load some context, or choose any action to build a different kind of run.
Generated pseudocode
VB-ishProgram loaded. Waiting for a human decision.
What the parts mean
A division of
cognitive labor.
Substantial AI contribution and final human judgment aren’t opposites. They’re different jobs in the same environment.
AI can
retrieve · mix · compare · propose · transform · challenge · draft · model · execute
Fletcher must
recognize · reject · correct · protect distinctions · verify history · decide what rings · decide when the Form is faithful · decide when we’re done
The Field
Files, conversations, lived experience, unfinished ideas, artifacts, taste, and facts retained long enough to become available again.
Cheap collisions
I can hand AI A, B, C, D, E, and F. It shakes them together and asks, “Purple?” Usually that’s a candidate, not an answer.
“Hmmm. No, but…”
A wrong proposal gives me something concrete to push against. The push reveals a distinction. That correction becomes new material. Mix again.
Leverage after Ding
Once something rings, AI can investigate, challenge, draft, model, specify, build, and revise what rang.
History returns
The artifact returns to the Field. So does the collaboration itself: a searchable primary-source record that future archaeology can retrieve.
Saved output
Real runs.
These are compressed logs from actual work.
RUN_01ECHDING + CORRECTION + BUILD
ARCHAEOLOGIST recovered an older recruiting artifact and re-read it against newer evidence and systems thinking.
HUMAN CORRECTION The software should conduct the interviewer, not replace the human interviewer.
DING fixed methodology + variable company context + bounded generative AI → bespoke candidate exercise
RESULT AI participated heavily. The product thesis depended on the correction.
RUN_02The Way They See ItWEIRD QUESTION + OLD MATERIAL
AI QUESTION “What happens when an engineer observes like a poet?”
COLLISION The question met an old bus thought experiment I’d carried for years.
HUMAN RECOGNITION The collision mattered. A comparative media format followed.
RUN_03Resident InventorMANY PROPOSALS + HUMAN VOTE
LOAD_FIELD Eight invention histories. I knew they shared something but couldn’t compute every case at once.
AI proposed abstractions. I rejected those that flattened important differences. Then: “and sometimes two of them touch.”
DING THE FIELD → COLLISION → DING → FORM → REALITY → FIELD
MEET_REALITY Codex built it. The first diagram looked like a prescribed process. Human response: “Hmmmm.” The architecture changed.
RUN_04PorpoiseSPECIALIZED AI JOBS
Porpoise is a close relative of this program. It assigns AI different cognitive jobs and stances: archaeologist, mixer, challenger, drafter, modeler, spec writer, builder, archivist.