Claude Code built an AI series based on "Electronic": the build cost $21.7 via API
The author of the Telegram channel "Tekhnokafe" put together an AI series based on "The Adventures of Electronic," with Claude Code running the entire build on the Opus 5.5 model. At API prices, meaning pay-per-token, the work came to about $21.7. On a Max subscription, it used 2.2% of the weekly limit. GPT Image drew the frames, Veo 3.1 Lite animated the scenes, ElevenLabs produced the voices, and all other paid services together cost about $4. The takeaway: the orchestrating agent cost more than the video models, and the hardest part was the characters' lips.
AI-processed from TG: @tekhnocafe (Технокафе / Вайбкодинг / AI-агенты); edited by Hamidun News
Can an AI agent be assigned not just to write code, but to build a short film from start to finish, and what does that cost in real money? The authors of the Telegram channel "Tekhnokafe" launched a series that moves the plots of Soviet films into a world of AI agents, and published a detailed bill for the first episode. Claude Code ran the entire build on the Opus 5.5 model. At API prices, meaning if you pay for every processed token, it cost about $21.7. On a Max subscription, which has a weekly limit, it used 2.2% of that limit.
Plot and a lesson for those who write code with an agent
The first episode is based on "The Adventures of Electronic." Codex replaced the robot, and Claude played Professor Gromov. Everyone remembers the plot: a double did the schoolboy's homework, but he still had to go to the blackboard himself. It works the same way with an agent. The task is solved, the code runs, but you cannot explain how it works. Imagine the agent wrote you an authorization module, and in code review you're asked why the tokens are stored exactly this way.
The advice from the series is simple: ask the agent to explain what it did. Claude Code has a mode for this. Open /config, select Output style, and choose Learning. In this mode the agent leaves part of the work to you, so you don't become a passive observer.
What the video is made of and what each piece cost
GPT Image drew the frames through Codex on a ChatGPT subscription, so there is no separate bill for them. Veo 3.1 Lite animated them: six scenes were made on the Google Cloud trial credit, and two more were ordered through the kie.ai service at $0.15 each. Veo failed to deliver one scene, and Kling 2.6 made it for $3.35. Another $0.20 went to an unsuccessful Seedance test. ElevenLabs voiced the characters for less than $0.2. LatentSync synchronized the lips, and the 1970s film grain and music were assembled locally with code.
Adding up what cost money at list price gives about $26: $21.7 for the agent's work and about $4 for the video, voices, and the failed test. The ChatGPT subscription and the free Google credit are not included. So the biggest expense line is the orchestrator itself, which wrote the pipeline code and managed the services, not video generation. The video models themselves cost only a few dollars for a short clip like this.
What turned out to be hard
The video contains no frames from the film and no actors' faces; the world was rebuilt from scratch: a courtyard with swings and cottonwood fluff, a classroom with a chalkboard, a lab with reels. The costumes were checked against the film, and each frame has its own background extras.
The hardest part was not the picture but the lips. The scenes had to be redone several times so the characters actually spoke the words. A video model easily produces a beautiful frame, but matching speech requires a separate tool and several iterations. The second lesson concerns refusals: when Veo rejected a scene, another model covered it for a couple of dollars. A pipeline where one generator can be swapped for another proved more practical than betting on a single model.
Caveats and what practitioners should take from this
This is the author's personal post, and the figures have not been independently verified. The cost does not include the person's time, subscriptions, or free credits. Judging by the description, the video is short, so the calculation cannot be extended to a full-length series.
For product managers and founders, the takeaway is this: a short video can be made with an agent and several models for tens of dollars, if orchestration is handed to the agent. Quality control costs the most, above all speech synchronization. For developers, the takeaway is different: turn on Learning mode and demand explanations, so you understand what you are responsible for.
The next episode comes out on Friday, based on "Kin-dza-dza." The senior orders an anti-gravity device through a chain of "pacaks," the fifth retells it to the fourth, and in the end they bring a wheel. The authors promise to analyze where a chain of subagents loses context and when a single "ku" is enough. The topic is practical for anyone building multi-agent systems: every retelling between agents can distort the original task.
- Nov 21, 202622 Claude Agents Spent Six Weeks on Black Hole Problems — and Solved Two
- Nov 21, 2026Google to split Gemini chat history from model training, so you can opt out without losing your history
- Nov 21, 2026Anthropic Goes Public: Up to $64.4B Raise, Ticker ANT, Listing November 19
- Nov 21, 2026Telegram Patches HTML Widget Holes as QuickJS Audit Finds 27 Bugs
Key Questions Answered
What are the key details on Plot and a lesson for those who write code with an agent?
The first episode is based on "The Adventures of Electronic." Codex replaced the robot, and Claude played Professor Gromov. Everyone remembers the plot: a double did the schoolboy's homework, but he still had to go to the blackboard himself. It works the same way with an agent. The task is solved, the code runs, but you cannot explain how it works. Imagine the agent wrote you an authorization module, and in code review you're asked why the tokens are stored exactly this way.
What the video is made of and what each piece cost?
GPT Image drew the frames through Codex on a ChatGPT subscription, so there is no separate bill for them. Veo 3.1 Lite animated them: six scenes were made on the Google Cloud trial credit, and two more were ordered through the kie.ai service at $0.15 each. Veo failed to deliver one scene, and Kling 2.6 made it for $3.35. Another $0.20 went to an unsuccessful Seedance test. ElevenLabs voiced the characters for less than $0.2.
What turned out to be hard?
The video contains no frames from the film and no actors' faces; the world was rebuilt from scratch: a courtyard with swings and cottonwood fluff, a classroom with a chalkboard, a lab with reels. The costumes were checked against the film, and each frame has its own background extras. The hardest part was not the picture but the lips. The scenes had to be redone several times so the characters actually spoke the words.
Build & Ship Production AI Agents Without Boilerplate Code
Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.