Deep codebase context cuts Claude Code's token cost by 47% (bito.ai)

Bito has announced a significant enhancement to its Claude Code AI coding agent with the introduction of AI Architect, which reduces token costs by 47% on the SWE-Bench Pro benchmark. This innovative system provides the coding agent with an up-to-date, structured map of the entire codebase, allowing it to bypass extensive exploratory steps. Traditionally, coding agents would exhaustively search through files and imports to understand unfamiliar repositories, incurring high token usage. AI Architect streamlines this process, enabling the agent to directly access relevant files, resulting in a 68% reduction in costs for individual tasks and 60% fewer reasoning steps overall. The implications for the AI/ML community are substantial, as this development not only enhances cost efficiency but also improves task success rates from 51.9% to 70.1%. By continuously indexing repositories and providing targeted queries during execution, AI Architect minimizes unnecessary navigation and reduces redundant operations like file readings and shell commands. For engineering teams, this translates into substantial savings and productivity gains, especially in larger, more complex codebases, without requiring changes to existing workflows or AI models.