#aiagent Community Insights & Market Sentiment

Chess's Grok Bot Diary #3 🎬

【Bot Updates】
• Save login info, bank cards, and contact details: Bot can now save your login information, bank cards, and contact details. Add them once, and Bot will securely remember them for direct use later.
• Mobile app integration: Bot can now use apps on your phone such as Calendar, Reminders, and Contacts, with iOS and Android rolling out gradually.
• 1Password: Bot can use your 1Password account. Once you approve the logins you want to share, Bot will log in for you, and access will roll out to users over the next few days.
• Pages: Bot can make reports, memos, and dashboards. Ask Bot for a page, and you can edit it together with Bot, or let Bot keep updating it continuously.

【Chess's Case Study】
This episode tells a complete story: on October 24, my Bot "Caesar" and I turned one of the hottest videos on X into a Bagel version in a single day: "Checking my @bagel_win account 10,000 times." The finished piece was made entirely by Caesar from start to finish; I only gave feedback, going back and forth more than a dozen times. The whole process happened on my phone, chatting back and forth with my personal house manager Caesar.

1️⃣ 12:00 Requesting the task: I dropped a reference video in the group and said, "Based on similar videos, and the Notion one from earlier, make a Bagel video." At 12:14 I sent the Coinbase one, "Checking my portfolio 10,000 times." First use my X avatar, and make it about checking the @bagel_win account 10,000 times. PR was also breaking down the shots, and Caesar said it would handle the finished piece.

2️⃣ v1–v4, AI generated the video directly: v1 was a 10-second avatar version, with a white room and frantic folding open-and-close action, but the hands and phone warped, and the mouth opened too wide at the end. I asked for a Bagel logo on the back of the phone, the main character to be Caesar's avatar, and a yellow background to contrast the white Duo. I also asked it to follow the original post with head movement, the entire Duo flipping over, and a night-city background. In v3 the image broke at 0:07, and Caesar's self-check told me, "Don't use this version yet." In v4 the phone only shook and didn't open or close; it failed self-check, so it didn't send it to me, and concluded: AI-generated video can't produce real hinge opening and closing.

3️⃣ v5, a new approach: at the same angle, first generate two stills, "closed" and "open". Caesar wrote the script itself, used ffmpeg to alternate them into cuts at about 4 opens/closes per second, with 5 camera angles hard-cut every 2 seconds, plus synchronized clicking sounds. I said, "This version is pretty good." Small twist: the backend self-check said it couldn't see the opening and closing clearly, so Caesar reviewed frame by frame and found that frame extraction had missed it, and honestly told me, "Go by what you can see."

4️⃣ v6–v9, polishing the details: some parts of the Duo turned into 4 squares, and they had to match the real device exactly; the inside of the Bagel had to be filled with yellow; the night-city background didn't look good, so we switched to a pure yellow background, and I sent a yellow-background white-face logo; add the Bagel text under the logo on the back; use Caesar avatar's color value #FFFF01 for the yellow background, and add the small vertical line from the avatar. Caesar also proactively reported issues, like the AI writing the sign as "CAESR." By v9, I sent it the official Duo assets, and it redrew the white glass back panel and the capsule dual cameras based on them.

5️⃣ The biggest creative idea: turning "10,000 times" into "10,000 emotions." The 5 shots were, in order: nervous, angry, aggrieved, ecstatic, and collapsing into laughter. v10 added large Chinese text on the left of the avatar, with the English tag underneath.

6️⃣ v11–v14, director-level details: for "nervous," it had to wait 0.7 seconds after the phone started shaking before popping up; when nervous, the eyes weren't looking at the phone, so it redrew it; before nervous, it had to first smile with anticipation for 0.8 seconds; at the end, add 1 second of black screen with the Bagel logo, text, and slogan. At 17:41 I said, "Not bad," and locked it in.

7️⃣ v15 voiceover: I wanted the final 1 second, when Bagel appears, to have an American male voice with an ad-like tone, saying Bagel and the slogan together. Caesar kicked off a background task, found an existing TTS voice (GuyNeural), and had it say, "Bagel! A mobile-first prediction market app." It was about 2.1 seconds long; the 1-second end card couldn't fit it, so it extended it to 2.4 seconds and told me honestly. I said, "This version is fine," deleted the old post, and reposted it with sound.

The whole process: request at 12:14, final version at 17:41, voice added at 18:05, published at 18:10 — 15 versions in total. It can break down reference videos, extract avatars, switch approaches when video generation doesn't work, use image generation plus scripting for editing, add click sounds and TTS voiceovers, run self-checks before sending each version, and honestly report errors. I only have to say what I want; that's how I feel about Grok Bot now: a partner on my phone that I can talk to and get real work done.

The image is a nine-grid selection from the 15 versions, and the last three images are screenshots of me chatting with Caesar on my phone. The finished video is in my X account @chessxyz's October 24 post; no link is provided here.

This content is a record of a personal AI Agent usage experience; both the video and text involved AI participation.

#GrokBot #AIAgent

"Love, Death & Robots"🩷☠️🤖

That's the end of this episode!

aiagent

Trending Topics