How to use Replit Agent to build a fighting game
Replit Agent and Claude Code built a podcast-inspired fighting game. The maker wrote story arcs, reviewed consistent character poses and tuned the experience.
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Implement the combat loop and screens from these fighter and story specifications using React and TypeScript. Add clear hit and combo feedback, preserve character identity across supplied pose assets and let me review story, art and balance before publishing the playable game.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- Andrey Esipov
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- The Operators game was published with agent-implemented combat and screens plus human-authored story arcs and reviewed assets. Other models supplied media; the agents did not make every creative decision.
- Human involvement
- Esipov authored story arcs, reviewed character images and animations and tuned combat, levels and dialogue.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Esipov defined podcast-inspired fighters and story arcs while the agents implemented combat and screens.
- Character assets were generated from a consistent idle-pose reference and reviewed by the author.
- The author tuned combat feedback, dialogue and levels before publishing the playable project.
Andrey Esipov built Operators, a fighting game based on guests from a podcast archive, using Replit Agent and Claude Code for implementation. A product manager rather than a professional game developer, he worked through the buildathon project over two weekends.
The agents implemented combat logic and screens with React, TypeScript and Zustand. Framer Motion supplied visible feedback such as hit flashes, screen shake and combinations. The game connected podcast ideas to fighting mechanics instead of simply turning episode summaries into a static quiz.
Asset consistency was a separate challenge. Esipov first generated an idle pixel-art pose for each fighter, then reused that image as a reference for additional poses so faces and outfits stayed recognizable. He reviewed repeated generations and authored the story arcs himself. Music and character dialogue also involved other generation tools, with human choices about where and how they appeared.
The published game and its described combat loop are the concrete result. The source does not support attributing every story decision or asset choice to the coding agents. Esipov's account shows an agent-assisted implementation combined with substantial editorial and visual work, where human review of recognizable characters and tuning of the experience mattered as much as getting the screens to run.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Substack · Andrey Esipov
Frequently asked questions
What did the coding agents implement?
Combat logic and game screens using the web application stack.
How were character poses kept consistent?
Additional poses used each fighter's initial idle image as a reference.
Who wrote the stories?
Esipov authored the story arcs and made editorial and balance decisions.
More details
About this account
Esipov defined podcast-inspired fighters and story arcs while the agents implemented combat and screens. Character assets were generated from a consistent idle-pose reference and reviewed by the author.
First-person buildathon account opened. Code involved both Replit Agent and Claude Code; visual, music and voice assets used additional models. Competition and timing claims remain the author's report. Suggested prompt written by Agents In Use; no original task prompt was published in the verified account.
Credit: Andrey Esipov — Substack
Source published
Added to Agents In Use
Source checked
Tools and topics
Replit Agent · Claude Code · React · TypeScript · Zustand · Framer Motion
Fighting game · Asset consistency
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.