Why demo videos rot, and what we did about it
Savas Vedova4 min read
Every demo video has a shelf life, and it is shorter than anyone plans for.
You record it on a Tuesday. The flow is clean, the cursor goes where you meant it to, nobody messages you halfway through. It goes on the landing page, in the onboarding email, in the sales deck. Then somebody renames New invoice to Create invoice, the settings move behind a tab, and the video is now a recording of a product that does not exist. It is still on the landing page. It will be there in six months.
The reason is not that people are careless. It is that re-recording is expensive in exactly the way that makes it get postponed forever.
What re-recording actually costs
Doing it by hand means: clearing your desktop of anything you would not want in a frame. Turning off notifications. Finding an account with data that looks real enough. Getting to the starting screen. Recording. Fluffing a click at 0:40 and starting again. Fluffing a different click at 1:10 and starting again. Trimming the dead air, writing the narration, recording the narration, discovering your voice sounds like that, recording it again.
That is an afternoon for ninety seconds of video. It is worth doing once. It is never worth doing because a button changed its label — so it does not get done, and the gap between the video and the product widens every sprint until the video is quietly deleted rather than replaced.
The interesting part is that none of that cost is in the demo. It is in the mechanics of recording one. The demo itself — what you show, in what order, to prove which point — is the ten minutes of thinking you did before you pressed record.
Separating the demo from the recording
So we split them.
The demo is a plan: a handful of scenes, each one a claim and the screens that prove it. It is text. You can read it in twenty seconds, argue with it, drop a scene, reorder two, and hand it back.
The recording is what a machine should do. Given a plan, driving a browser to the right screen and clicking the right things is a mechanical job with no judgement in it, and it is exactly the sort of job that does not get bored on the fourth take.
Corgy is that split. You describe the demo to your coding agent, in the repository of the thing you are demoing. It reads the router, the pages, the labels as they are actually rendered, and comes back with the plan — not a video, a plan. When it looks right, the Mac app drives your real cursor across your real screen, signed into your real app, and records the run.
Nothing about that is a rendering of your product. It is your product, on your machine, with your data.
The bit that surprised us
We assumed the hard part would be the driving. It was not. The hard part is knowing what to aim at.
An agent planning a demo from screenshots is guessing: it sees a button, it
guesses the label, and half the time the guess is close enough to sound right
and wrong enough to miss. An agent planning from your codebase is not guessing.
It knows the route is /billing/invoices because the router says so, and it
knows the button says Create invoice because that string is in the file it
just read.
That turns out to be the whole difference between a demo that records first time and one that stalls on step four. It is also why the planner lives where your code is, rather than in a web app looking at a screenshot you uploaded.
One claim, proved on screen
The other thing we keep relearning, and the one piece of this that is worth taking even if you never touch Corgy:
A good demo makes one claim and proves it on screen, in under ninety seconds.
Most demo videos fail before any of the recording mechanics matter, because they are a tour. Here is the dashboard. Here is settings. Here is the sidebar. A tour has no claim in it, so there is nothing for the viewer to be convinced of, and they close the tab at forty seconds having learned that your product has a sidebar.
Pick the one thing your product does that a competitor's does not. Show somebody doing that thing, start to finish, including the boring middle. Stop. If there is a second claim, that is a second video.
Constraining a demo to one claim also happens to make it cheap to re-record — which is the point. A ninety-second video with one idea in it can be regenerated the week the button gets renamed, because regenerating it is a sentence and a few minutes rather than an afternoon.
That is the version of this we want: a demo that is as easy to update as the screenshot in your README, and therefore actually gets updated.
If you want to try it on your own product, the setup guide is six steps and the first four are one-time. The free tier is enough to plan, record and narrate a first demo end to end.
Want one of these for your product?
Describe the demo in a sentence. Corgy records it on your own screen and writes the narration.