Yesterday we released a fully AI-generated ad we've been working on for the past few weeks. As far as AI slop goes, we're honestly pretty surprised and happy with how this came out.
We used around 5 different AI tools to make this happen, from generating shots to cleaning things up and stitching the final piece together. As you'd expect, continuity was a major issue, but we found some pretty effective solutions.
Yup, it's not the greatest solution for teams as a central issue tracker.
The main reason I included the mention of small teams is that you can opt out of committing ./ghist, keeping it on your machine only and use it to track your own set of work across multiple agent sessions. It then becomes more of a disposable tool you can use to get big chunks of work done with persistent task memory.
Going with sqlite might not have been the best decision either as it's ultimately a binary file that can't be diff'd. Potentially JSON might have been a better solution for this.
We used around 5 different AI tools to make this happen, from generating shots to cleaning things up and stitching the final piece together. As you'd expect, continuity was a major issue, but we found some pretty effective solutions.