AI-generated tutorial videos: one line in, a finished tutorial out
We typed one line into codus: "Create a tutorial on how to set up a free account on Olivia Brown AI." The agent had never seen the product. Two hours later there was a finished tutorial video on disk: filmed, narrated, animated, and quality checked. Nobody touched the computer in between.
How the AI agent made the tutorial video
The agent ran the whole playbook on its own, and every step below is ordinary shipped machinery rather than a staged demo.
It googled the brand, cross-checked the results, and verified it had the official site before doing anything else.
It created its own account. Every codus install carries an email inbox for its agents, so the verification code landed in the agent’s own mail and it typed the code itself. It has a real mobile number for SMS codes when a product asks for one.
It screen-recorded the signup flow in a browser session captured tab-only, with a visible cursor that glides, hovers and clicks at human pace.
It wrote the narration as a script first, in plain read-aloud language, then generated the voiceover from it.
It added the graphics a tutorial actually needs, click rings from the real clicks and step callouts, under our rule that every graphic has to earn its place. Every sound effect carries its licence in a ledger or the build fails.
An independent QC agent re-checked the final render: a verbatim pass on the finished audio, sync and framing gates, and a real watch-through like a viewer.
Why the footage looks human
The filming engine drives a real browser and draws its own cursor, because synthetic input never moves a system pointer. Movements are eased over most of a second, typing lands at seven to nine characters per second with natural jitter, and scrolls take the time a person would take. The capture is the browser tab and nothing else, so a stray notification can never leak into a take. If a step fails, the engine assembles the footage it already has and drops a screenshot of where it stopped, so a broken run still salvages its material.
The economics of AI-generated tutorials
There are no per-video fees from anyone in this pipeline. The agents run on your own Claude subscription, the voice runs on your own ElevenLabs key, and codus never meters usage. That changes what you can afford: we are queueing a how-to for every feature across our own products, because at these economics a library of two hundred tutorials costs what the subscriptions cost.
Try it
The tutorial factory ships inside the Video discipline, a per-seat module on top of codus. Brief an agent once, go do something else, and come back to a finished video. The full story of the pipeline lives on the Video page.
See it instead of reading about it.
Codus is out now on macOS.