ta-amps

Video storyboards

Storyboard draft 1: two videos

The first storyboard, kept as history. It planned two videos. The current plan replaces both with one nonprofit story.

Draft 1, October 29, 2026, afternoon. Kept as it was written.

How to read it

  • Real It worked when this draft was written and could be filmed as it was.
  • Building Someone was building it. It was not ready to film.
  • Needed It did not exist yet.

Video 1: AI is smart. The way we use it is frustratingly dumb.

A developer asks a coding agent for a video made with code. On its own it stalls; connected, it finds fresh tools, and the work splits into small jobs many workers share. Target 2:30 to 3:00, wide 16:9 plus a tall 9:16 cut.

  1. 0:00 to 0:12

    On screen
    The title as motion graphics: the mosaic builds tile by tile; “smart” lights blue, “frustratingly dumb” lights orange.
    Narration
    “AI is smart. The way we use it is frustratingly dumb.”
    Status
    BuildingThe scene recipe in the video seed pack.

  2. 0:12 to 0:35

    On screen
    Three cards animate in, then the line “Training doesn’t fix it” slides under them.
    Narration
    “It doesn’t know what came out an hour ago. It rewrites code that already exists, because a model is compressed knowledge, not a code library. It needs babysitting on long jobs. And training doesn’t fix any of that.”
    Status
    Building

  3. 0:35 to 0:55

    On screen
    Real screen capture: a developer asks their own coding agent, “Make a 30-second product teaser with motion graphics as code.” It answers with basic video generation, outdated or paid tools, and no motion graphics as code.
    Narration
    “Here’s a real request. The agent means well, but it’s stuck in the past and starts from scratch.”
    Status
    RealCapture any agent with no connection.

  4. 0:55 to 1:10

    On screen
    Real capture: they copy the starting brief from home.ta-amps.workers.dev and paste it into the agent.
    Narration
    “So they connect it. One copied prompt.”
    Status
    Real

  5. 1:10 to 1:35

    On screen
    Real capture: the same request now comes back with recent nouns, such as a repo published eight hours ago and its benchmark, each with a timestamp and a source.
    Narration
    “Now the shim puts what other people already found in front of the agent: tools published hours ago, with benchmarks on our hardware.”
    Status
    NeededA tool-discovery run that stores tool and benchmark nouns, plus a Herma shim rule for video tasks.

  6. 1:35 to 2:00

    On screen
    Real capture of the Tessera board: no reusable caption component exists, so the request splits into linked verbs (research, build the component, render, check). Several agents claim them, and a progress link appears.
    Narration
    “Nothing reusable for animated captions? It breaks the job into small tracked tasks, and a swarm of workers builds them. Anyone can follow along, or pick it up.”
    Status
    RealThe board and the task graph.BuildingThe video verbs.

  7. 2:00 to 2:15

    On screen
    The learner view: which models and verbs worked, plus a thumbs-up stored as a feedback noun.
    Narration
    “It learns which models and steps work, and it listens to feedback.”
    Status
    RealThe learners are merged code.NeededFeedback nouns and a way to give them.

  8. 2:15 to 2:40

    On screen
    The reveal: this video’s own verb graph. Script, storyboard, audio layers (narration, music, effects), video layers (motion graphics, captures, captions), render, frame check and review, with tool nouns updated a day earlier.
    Narration
    “And that’s exactly how this video was made.”
    Status
    BuildingThe video pipeline.NeededRunning the production as verbs, so this line is true.

  9. 2:40 to 3:00

    On screen
    Close: “Point your agent at it”, with links.
    Narration
    “Try it: home.ta-amps.workers.dev.”
    Status
    Real

The making-of cut

A 60-second cut shows the Tessera event log of this video’s verbs, with every asset’s hash, and the human narration verb highlighted.

What must exist before recording

In order:

  1. The video seed pack renders, encodes and checks wide MP4 files. This pipeline is being built.
  2. One real tool-discovery run that stores recent repo and benchmark nouns. The seed packs for it are being built.
  3. A Herma shim rule for video-as-code requests that returns those nouns with their dates.
  4. The production itself queued as a graph of verbs on Tessera, so the board capture is genuine.
  5. Feedback nouns and a simple thumbs control, as in the feedback design.
  6. Taylor’s narration recorded, as a person’s verb, with captions from Voice Transcribe.
  7. Music and sound effects with checked licenses (public domain or owned), stored as nouns with their license details.

Video 2: Spare tokens build tools for good

A volunteer lends their agent a few spare minutes. Through Eranos it builds a small, tested kit for a nonprofit; another volunteer’s agent checks it, and a person signs off. About two minutes.

  1. 0:00 to 0:15

    On screen
    Two facts, each needing a source before use: many people pay for AI they don’t fully use, and many nonprofits can’t afford the tools they need.
    Narration
    “What if your unused AI could build tools for people who need them?”
    Status
    NeededSources for both facts, or the facts dropped.

  2. 0:15 to 0:40

    On screen
    Real capture: a volunteer copies the starting brief from the Eranos site into their agent.
    Narration
    “A volunteer connects their own agent, with tokens they weren’t going to use.”
    Status
    RealThe site and its brief.

  3. 0:40 to 1:10

    On screen
    The agent claims a brief, “Intake checklist for a legal-aid hotline”. It builds the kit (a skill file, a prompt and synthetic test cases) and checks its own work.
    Narration
    “It picks a real need and builds a small kit, with tests.”
    Status
    RealFive example kits, on a local machine.NeededA hosted run, captured live.

  4. 1:10 to 1:30

    On screen
    A different volunteer’s agent reviews the kit (a builder can’t review their own), then a person releases it.
    Narration
    “A different volunteer’s agent checks it, and a person signs off.”
    Status
    RealThe review rules are enforced.

  5. 1:30 to 1:50

    On screen
    A nonprofit downloads the reviewed kit. On screen: “People keep every decision about a person.”
    Narration
    “The nonprofit gets a tested tool, and people stay in charge.”
    Status
    Real

  6. 1:50 to 2:00

    On screen
    Close: the Eranos link and “Donate your spare tokens”.
    Narration
    “Lend your agent a few minutes.”
    Status
    Real

What must exist before recording

  1. A hosted volunteer run captured from start to finish: the building agent, the reviewing agent and the steward’s release.
  2. A source for every statistic shown, or the statistics dropped.
  3. A short, plain preview page for a kit.