TechPrufer
Sign inGet started

Blogs›Baseten›Agentic Code Transformation and Model Orchestration

Agentic Code Transformation and Model Orchestration

Agentic Code Transformation and Model Orchestration

1
posts
2026

This post details the use of Laguna S 2.1, an agentic coding model, to perform a repository-scale transformation of the Hypersomnia game. It highlights Laguna's ability to understand an unfamiliar codebase, coordinate changes across multiple files and specialized AI models (Step 3.7 Flash, Krea 2 Turbo, NVIDIA Cosmos 3 Nano), and continuously verify its work. The process involved transforming game assets, code, and text while preserving core gameplay mechanics. Laguna demonstrated judgment in selecting and orchestrating models, scaling generated assets, and making tradeoffs to ensure a stable, playable output. The capability extends to complex, long-horizon tasks requiring creative problem-solving and technical constraint adherence.

2026

Laguna S 2.1 goes Greek: a repository-scale game transformation

8/3/2026

This post introduces and demonstrates the application of Laguna S 2.1 for a repository-scale game transformation. It details how Laguna orchestrated specialized models (Step 3.7 Flash for language adaptation, Krea 2 Turbo for asset generation) to change a game's aesthetic from cyberpunk to Ancient Greek. Key technical aspects include managing a 1 million token context window, coordinating changes across hundreds of sprites and dozens of code files, compositing generated assets while preserving dimensions and transparency, and making strategic decisions about which models to use (e.g., not using Cosmos 3 Nano for video generation within the game itself). The post quantifies the effort in terms of time, tokens, model calls, and modified assets, showcasing the agent's ability to handle complex, creative, and technically constrained tasks.