Asked for a video — got a department: Claude and 161 Codex sessions

What the author said

In publications AIstify and a AgentPlays the source was named as a post by Ananth, @Ananth7e on X, from October 14, 2026.

According to the text there, he asked Claude Code to recreate a 63-second video in a specific animation style. The assistant noticed the installed Codex and decided to involve it for image generation. The author writes about launching 161 sessions; Claude supposedly monitored memory, handled failures, and gathered the result.

The screenshot shared along with the post shows a plan for 161 sessions and 700 frames. Even in this plan, there is a common production problem: tasks need to be redistributed because some sessions only get one or two frames.

Where Telegram added drama

What we couldn't verify

During verification on October 20, the original X page was not open for direct reading. We cross-referenced the submitted screenshot with available reports. They refer to a single author, so they are not independent confirmations of execution.

AgentPlays notes that at the time of its publication, the video was still in progress. In the reviewed sources, we did not find a complete launch log, a verified final video, or actual costs. This does not prove that there is no result; it is the limit of our verification.

The most interesting thing here is not the number 161

For the reader trying Vibe Coding, this story has a useful question: what exactly counts as a completed task? "Many processes launched" sounds impressive. "Received a video that can be watched and evaluated" is a verifiable result.

Let's imagine two scenarios. In the first, dozens of attempts were made, but the character's face changes between frames. In the second, there are fewer attempts, but the motion is consistent and the file opens. For a client, the second result has value. This is our hypothetical example, not a description of Ananth's video.

Therefore, before a large-scale generation, it is worth agreeing on three things:

• Small sample: first, test one short scene.
• Boundaries: which tools are allowed and what budget cannot be exceeded.
• Acceptance: where the finished file is and what criteria define it as successful.

This is editorial advice for your own experiments. It does not mean that the described approach was reproduced or that a specific tool is guaranteed to meet all constraints. Also read about verifying results in our AI response verification guide.

But the increase is still good

The meme version of this story lives by office laws: as soon as work appeared, someone eager to distribute it was found. But for now, it's better to leave the title "manager of 161 sessions" as a joke. A serious conclusion requires results and cost figures.

A shareable formulation: the user reported that Claude Code involved 161 Codex sessions in working on the video. The screenshot shows a plan; we did not receive independent confirmation of full completion.