Day One

What you’re watching

The film explains the idea this whole site is built on. An AI is like a brilliant new hire on their first day: it has read more than anyone you know, and it knows nothing about your company. Give it a one-line task and you get a generic email full of placeholders. Give it the same task with four short pieces of context (who it’s for, why the meeting moved, what went wrong last time, an email you liked) and the draft rewrites itself, line by line, into something specific.

Then one sentence lights up orange: “I’ve updated the invite.” Nobody told the AI that. It guessed. So the film tells you to check before you send, then ends on the point the rest of this site makes at length: when the answer is generic, it usually says more about the brief than about the model. The long version is in Your AI Has No Idea What Your Company Does.

Two prompts, then a question about sound

Claude Motion went into beta on 8 October 2026. Claude builds the animation as code (shapes, text and timing) instead of generating footage, so every word and every second stays editable. You watch it in a player, pin comments to a moment, and export an MP4.

I wasn’t hands-on at all. In the first prompt, I asked Claude Code, which works directly on this site’s files, to draft a few video ideas for the website. It came back with a plan: three short explainers built around the brilliant new hire, a ranked list of 21 more ideas, rules for the series (every word on screen, the last frame has to work as a still image, no model names) and a storyboard for each, timed to the second.

In the second prompt, I asked whether Motion works in Claude Code and told it to make the first idea. It built the film scene by scene, seven scenes in 44 seconds, in this site’s colours and fonts. This project’s rules say every public word gets a separate review, so a second agent went over every line on screen. That review changed five lines and one detail of the picture. The single word “Check.” became “Check before you send.”, because on its own it could read as “all checked”. And the meter showing how well the email fits stops at three quarters, because a full meter right before the guess would contradict it.

Then I asked whether it could add sound, a voiceover for example. A voiceover isn’t possible: Motion has no voice and doesn’t accept audio files. What it can do is compose music and sound effects timed to the picture. So that’s what it added: footsteps as the new hire walks in, a bell for each icon that appears, a soft thud when the sticky note lands.

Claude never looked at its own film while making it. The instructions that come with Motion tell it not to render frames unless you ask, so the first person to watch it was me. I found nothing to fix, and even the music works. For an experiment without a single round of feedback, it’s as good as it needs to be. I exported it at 1080p and 60 frames per second, and that’s the file playing above.

Why two prompts were enough

I’m amazed how simple this was, and how good the result is. But the two prompts aren’t the whole story. Claude Code works inside this site’s project, and the project already holds what a film like this needs: who the readers are, how I write, the colours and fonts, the brilliant new hire and the guides that explain it, and the rules for what the site never says. That’s the AI OS behind this site, the same kind of setup I describe for a company in How We Built an AI That Knows Our Company. Without it, I wouldn’t have got a result like this from two prompts. That’s what the film is about, too.

What it can’t do yet

• No voice. Motion makes music and sound effects. A voiceover has to be recorded separately and added in a video editor after export.
• Beta, and not on every plan. For now, Motion is only on Claude’s Team and Enterprise plans, and on Enterprise an admin has to switch it on.
• One film, one round. This was an experiment, and nobody gave it feedback. For real work, I’d put most of the effort into the foundation: what exactly has to come across, and to whom. Then I’d do a few rounds of improvements.

Where I’d use it

If I worked harder at it, I could see myself producing good videos for all sorts of things. Anywhere one or two core ideas need to come across in a way people remember. Learning moments and training sessions, to make them more entertaining. And as a product manager, I see it as a good way to introduce a new feature or explain a change in the product.

What you can learn from this

• Build the foundation first. Two prompts were enough because the project already knew the audience, the voice and the look.
• Let the AI plan before it builds. The storyboard from the first prompt is what the second prompt turned into a film, and a plan is quicker to review than a film.
• Have the words checked separately. A second pass over every line on screen changed five of them.
• Design for the sound off. Feeds like LinkedIn start videos muted. If every word is on screen, the music is a bonus.