What this is
A bilingual archive that observes, curates, and comments on AI-era artifacts: fake products, failed prototypes, absurd launches, and the tools people actually build with AI.
Curating failed prototypes, strange objects, and ideas worth salvaging. Every artifact is reviewed, numbered, and labeled.

A CANVAS FOR THOUGHT BEFORE THE PROMPT
Let an idea keep forming through space, action, language, and time.
This is the first time I am introducing Canvas Prompt on the Institute's own site. It keeps the process before compression in view, then lets that context enter an AI workflow. The Prompt is not the beginning.
Recent observations on agent quotas, synthetic voiceover review, and how public reset signals enter real work.

01 · Curated Observation · Agent Infrastructure
Cloud deployment is old. An agent control plane that can route work across devices, sandboxes, and workers changes the structure.
Read the post →
02 · Curated Observation · Agent Economics
Quota rules are becoming part of the way people decide what work an agent receives today.
Read the post →
03 · Tool Release · AI Video Workflow
Review the generated take cue by cue, then hand the next pass something concrete.
Read the post →Field notes, observations, and working notes. Newest accession first.
AI is extraordinarily good at producing meaning. People still have to decide what deserves entry into their lives and when it has gone far enough.
When several agents share a project, the final output often survives. The reasoning that produced it is much easier to lose.
A public post becomes a window. A window becomes a task with a deadline.
Execution is being broken into smaller, describable parts. The remaining work starts earlier.
When the deliverable is a judgment chain, an AI handoff needs more than chat history.
The task did not change. I became its router.
Every exhibit sits inside the institute's frame: a specimen label above, the object displayed, a curator's note below.
Built on a real open-source soccer vision pipeline, with a 58-second launch film and a crowdfunding page. The engineering is real; the product is not. The institute's test: can sufficient rigor make an expert hesitate?
Someone posted "can someone build this?" and two days later, someone did. A leaderboard that treats free dev tools as deities you can offer incense to. Absurd form, real feeling: developers genuinely owe these tools a debt. The merit value is collective reputation voting in monastery robes.
A public post from a product lead is usually swallowed by the timeline. CodexRadar gives it an opening and a closing state: the window opens when a reset is announced, then closes when completion is confirmed. Remaining quota becomes a task with a deadline. Our earlier observation asks how resets shape platform value, capacity, and work allocation. This specimen records the user-side answer: build a folk instrument that misses fewer windows.
A bilingual archive that observes, curates, and comments on AI-era artifacts: fake products, failed prototypes, absurd launches, and the tools people actually build with AI.
Three exhibits, a growing post stream, and an open-source skill for auditing meeting judgment.
Not a research institution, not a media outlet, not a shop. It's a shelf where AI-era things worth a second look get put on display.