A first cut from the work already happening

This is another look back, to 10 September, when Cindy was making curry puffs. I asked our AI agent to help record the preparation and turn the material into long and short videos. We wanted a way to show some of the work that happens before the food reaches a customer.

The agent watched the camera view and used OBS to capture short sections of useful steps. It kept a record of what it had captured and what was still missing. Cindy continued making the pastry; the camera could only show the part of the work happening inside its frame.

The available footage included kneading, flattening and repeatedly rolling the dough, then cutting spiral rounds and rolling the shells. I confirmed that the repeated rolling was part of creating the layers. That detail mattered: the edit needed to describe Cindy’s actual process, rather than guess what a familiar-looking movement meant.

There were gaps. We did not have a complete filling or crimping sequence, the cooking process or a cooked-product reveal. When I confirmed that the work had ended, I asked the agent to try a first cut with what we already had. We would look at the result before deciding whether more footage was needed.

The agent produced three local first cuts: a 36-second vertical version, a 90-second horizontal version and a 90-second vertical version. It added English captions, removed the on-site conversations and used a cream background to keep the hands and pastry visible without forcing every shot into a narrow crop. We left these cuts without music.

Those files were a starting point for review. They showed pastry preparation and uncooked curry puffs, so we kept that description clear rather than calling them a complete recipe video. I am not claiming these cuts were posted or brought in customers. What we did complete was a set of viewable edits from real footage, with the missing steps still acknowledged.