Researchers have developed CRAFT, a two-stage post-training method designed to improve the reliability and efficiency of enterprise coding agents. This approach first uses schema-stripped supervised fine-tuning to teach agents domain-structured plans and behaviors without needing exhaustive schema documentation in every prompt. The second stage employs execution-shaped reinforcement learning to refine tool selection, code quality, and consistency, particularly in multi-turn analytical tasks. Evaluations in an advertising analytics environment showed CRAFT significantly improved composite Agent Score, consistency, and multi-turn coherence while drastically reducing input token usage and schema-discovery loops compared to a baseline method. AI
IMPACT This research could lead to more reliable and efficient enterprise AI agents, reducing computational costs and improving multi-turn analytical capabilities.
RANK_REASON The cluster describes a new method presented in an academic paper for improving AI agents. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
