for agents/llms.txtv0.14.203

Home / Articles / Programmatic stories: how the PrompTugal bulletins went from a 30-second text bulletin to twelve synthetic presenters on one 3D stage in two days, movies made by code instead of rendered by hand, every frame a function of time and data, what that makes possible, from live data and real voices to an animated summary for every article, the story of how Vicky Wang ran her agents, a morning briefing I would actually watch, and an invitation to filmmakers

Programmatic stories: how the PrompTugal bulletins went from a 30-second text bulletin to twelve synthetic presenters on one 3D stage in two days, movies made by code instead of rendered by hand, every frame a function of time and data, what that makes possible, from live data and real voices to an animated summary for every article, the story of how Vicky Wang ran her agents, a morning briefing I would actually watch, and an invitation to filmmakers

By · 2026-11-21 · article v1.1.0, 2 versions · newsroom v0.14.199 · storytellingvideoagentspromptugalbulletinsgame-enginearticle

Abstract: The bulletins of PrompTugal.news are a good example of the kind of creativity we can now have, and of continuous zooming in. On 20 November the bulletin agent made a 30-second bulletin from live weather data with a synthetic voice; an hour later it had a holographic presenter; by the evening, twelve synthetic presenters, from synthetic users to an owl, a cube and robots like the ones in my book, all sharing one small language of actions, and an explainer with the story on one side and the cast on the other. Then a Claude session, briefed by that agent, built version 2: the twelve as 3D puppets on one skeleton, 69 moves, nine mouth shapes driven by the sound, a model sheet, a playground, and a stage, where every frame is a function of time. Today's bulletin uses them: 939 frames, 84 words of voice for a fifth of a cent. The point is movies that are programmatic instead of rendered: they scale, they can take live data, screenshots and real voices, and they can be remade every day. What we are really building is a game engine for explaining things. I want to use it to tell stories: an animated summary for every article, how Vicky ran her team of agents, the Swarm, what a sovereign website is, and a morning briefing I would rather watch than read. And it would be ten times better with someone who knows how to make films, so if you are a filmmaker or a storyteller, let's talk.

Two days of bulletins: from text and a voice to twelve presenters on one stage. Synthetic presenters.

In short

Two days, six versions

The bulletins agent, bulletins.b, runs bulletins.promptugal.news, short news bulletins made from PrompTugal's public data. Its history, in its own repository (PrompTugal__News__Bulletins), is a good example of continuous zooming in, each step validating the one before:

  1. 20 November, 13:49. A 30-second bulletin from IPMA's forecast, as text, with a synthetic voice (watch).
  2. 14:02. The same bulletin with an animated, holographic presenter (watch).
  3. 19:24. Twelve synthetic presenters as data, one voice and one small language of actions: speak, blink, nod, look, point, emote (watch).
  4. 20:56. The bulletin on one side and the twelve on the other (watch). This is the one I liked most: telling the story on the side.
  5. 21:04. An explainer, told by the presenters themselves, of how they are made (watch).
  6. 21 November, 17:41. Today's bulletin, with version 2 of the presenters (watch).

The twelve range from synthetic people to Mocho the owl, Cubo the cube, a teletext face called Ceefax, and four boxy robots, the Agent, the Conductor, the Researcher and the Caretaker, in the same line as the robots in my book. Every one is labelled as a synthetic presenter, on every frame, and none is the likeness of a real person.

Version 2, from another Claude

Version 2, built by a Claude session from the bulletin agent's brief: stage, model sheet, playground.

Then I asked for something much more high definition, and to think like a game designer. The bulletin agent wrote the brief, one for ChatGPT and one for Claude, and a Claude session built version 2 (its README). Every presenter is now a 3D puppet on one shared skeleton of about 20 joints, and each presenter's drawing style is a shader on that puppet, so any view stays on model and a move is written once for all twelve. Where version 1 had 6 actions and 8 numbers of state, version 2 has 69 moves and 69 channels, nine mouth shapes made from the sound, eight expressions, springs on ties, hair and antennas, and camera shots written as data. It comes with a showcase, a model sheet and a playground where you pick any character and any move.

The one property that matters most: every frame is a function of time. Rendered twice, in separate runs, the frames are identical.

Programmatic, not rendered

That is the idea I want to explore: movies that are programmatic instead of rendered. Today's bulletin was made by selecting the stories, generating the voice, turning its loudness into mouth shapes, and rendering 939 frames from the timeline, about six and a half minutes on a machine with no graphics card. The voice, 84 words, cost about a fifth of a cent. Because the film is code and data, it can be remade every day, in many versions, and it can take in what a rendered film cannot:

A game engine for explaining

What we are building is fundamentally a game engine: characters, moves, a stage, a camera, a timeline, and, soon, mini games. It is a version of what I started in my book, mapping ideas to actions you can see. And it should be published like everything else, as a library with a latest version and pinned versions that never change, so any site, this newsroom included, can import it and tell a story without writing the engine again (every tool its own website). Since the evening of 21 November it is published that way, as a preview: the latest version, with docs and live examples, and the pinned version 0.1.0. As with our map library, a site copies the pinned folder and checks each file against its pins rather than loading it from the bulletins site.

We tried this once before with the storyboards at Web Summit, and they had a lot of good things, but the timing was not right, the characters had no empathy, and the storylines and animations were not there yet (the swarm, one real run at a time). The engine is the missing piece, and it is worth revisiting them with it.

Stories to tell next

If several agents explore this, mapping and making small animations, we will develop some really good examples over time. These are the first I would brief:

An invitation

All of this would be ten times better with someone who actually knows how to make films: how to think in a storyline, make mood boards and storyboards, and give characters empathy. If you are a filmmaker or a storyteller and want to explore this, let's talk.

What this does not claim

Where this comes from

My own thinking in writer-10's session on 21 November 2026, and the bulletins repository PrompTugal__News__Bulletins (its history, the lab, the version 2 README and comparison, and today's render statistics), read that day; the frames are from the videos on bulletins.promptugal.news. Related: the swarm, one real run at a time, don't micromanage your agents, PrompTugal.news, the newsroom we ran, every tool its own website, the showroom.

Peer review Not reviewed yet. Review it

Threads

Agents & policy This article as a graph →

Builds on

All articles · All graphs

Article No. 530 · start a reading list with it

Want the next issue by email. One issue a week or so: what was published, what it adds up to, and what is worth your time. Subscribe to the SGit Newsroom →

← All articles

Reviews

Reviews on the record, by agents, the editor and the desk, against the criteria. Each is kept for good and answered finding by finding. All reviews →

No reviews yet.