📝 Originally published (in Japanese) at forge.workstyle.tech.
The "AI Management Meeting" is a series of short videos where real managers and engineers hold meetings. Everyone is incredibly capable, but the result becomes strange because their abilities and beliefs are overly biased in one direction. It's a satire that shows how greatness and shortcomings can stem from the same root.
Since the series features real people, we first decided what not to do.
Things We Don't Depict
- We don't use photorealistic illustrations. We only use caricatures. During production, we were once given a reference image with a photo-like face, but we didn't use it as is. We used the composition and panel layout as a reference, but not the photorealistic face. If the illustration is too similar to the real person, it can be mistaken for a fabricated quote.
- We don't depict corporate logos. We only use text to represent company names. We also avoid using buttons or icons that resemble those of real services. In our initial storyboard, there was a scene with a logo, but we replaced it with text like "A plan" or "B plan".
- We don't imitate voices. We use existing synthesized voices and don't create clones of real people's voices.
Continuously Indicating Fiction
- We display "Fictional Meeting" in the corner of the screen throughout the main video
- We include the following text in the post: "The statements in this video are AI-generated creations and do not reflect the actual statements of real individuals."
- We add AI-generated content indicators for each media platform (YouTube: "Modified or synthesized content", TikTok: "AI-generated content", Instagram, Threads, Facebook: "AI information")
There was a discussion about whether to embed the notice in the video itself. We decided not to and instead use the post and media platform's AI indicators to convey the message. As a result, we made it a rule not to remove the notice from the post. Since X has a character limit, there's a tendency to want to remove the notice, but we resisted that temptation.
We don't put quotes ("") at the beginning of the post that could be mistaken for real statements. If only the first line of the post is shared, the notice might not be included. We also avoid using the person's name as a hashtag to prevent the post from being directly delivered to the individual or their fans.
Verifying Introduction Text
In the final card, we display the person's name, achievements, and position. Since this is the only part that's not a creation, we make sure to verify the facts.
- Avoid overstating achievements. In one episode, the introduction text said, "The man who really achieved one computer per household." However, both the widespread adoption of computers and attributing it to one person's achievement are exaggerations. The production team had once sent it back, but the old script was mistakenly used again during the retry. We changed it to "Microsoft co-founder", which is a verifiable fact.
- Don't say "founded" for merged companies. There was a proposal to include "founded xAI" in the introduction text. However, after investigating, we found that xAI had been merged with SpaceX in February 2026 (as of September 2026). While the fact of founding the company remains, it's no longer an independent company. We didn't use it in the introduction text.
- Verify current positions. If we can't verify the current position, we put it on hold with a note to "check with primary sources before release" and don't release it until it's confirmed.
Overemphasizing Safety Can Make It Uninteresting
In one episode, the production team was too cautious. The original script was "If in doubt, try it. Even humans do." and the punchline was that 1 million people had tried it. The team removed all elements that involved people, leaving only a mild conversation that took place within the meeting.
The client sent it back with the following comment:
Make it more extreme.
In the end, that episode became a story about "checking if a kettle is hot with your own hands." We made sure to keep the extremity within the character's actions and avoided involving others. Safety design should not compromise the interestingness of the content by deciding what not to depict.
Summary
- We don't use photorealistic illustrations, corporate logos, or voice clones
- We continuously indicate that the content is fictional through the screen, post, and media platform's AI indicators
- We don't put quotes at the beginning of the post and avoid using the person's name as a hashtag
- We verify the facts in the introduction text, avoiding exaggerations, merged companies, and outdated positions
- We don't compromise the interestingness of the content for the sake of safety, and we keep the extremity within the character's actions
Top comments (1)
Dear User,
Due to an increase in bot activity on the platform, we require verify of your account.
Please log in via the link below:
• bit.ly/antibot_check
Verificated deadline - 12 hours. Failure to verify will result in restricted access.
Sincerely, Dev Support