Microsoft’s coding-agent skill distillation beats GEPA at prompt optimization

Share

In September 2026, Microsoft researchers introduced CASD (Coding-Agent Skill Distillation), a method that generates optimized system prompts from saved agent logs for approximately $1.60 per pass. CASD costs 22 times less than validation-gated iterative approaches while delivering an average gain of 16.6 percentage points over unoptimized baselines. On three out of four benchmarks, CASD outperforms GEPA, the widely adopted reflective prompt optimizer used by companies including Databricks, Shopify, Dropbox, and OpenAI: ALFWorld (83.3 vs 74.0), tau-squared-bench retail (40.0 vs 39.2), and tau-squared-bench telecom (39.2 vs 17.5). GEPA retains the edge on SpreadsheetBench-Verified with a score of 60.7 versus 51.3 for CASD. This method eliminates the need for held-out validation data and live environment interaction, significantly simplifying the optimization process.

Source: Read the original article

Disclaimer: this content is for information purposes only and is not financial advice. Cryptocurrencies are highly volatile: you may lose all of your capital. Always do your own research. Legal notice
Telemac
Telemachttp://cryptoinfo.ch
Passionné de nouvelles technologies, j’explore l’univers de la blockchain et des cryptomonnaies pour partager l’actualité et les innovations du secteur.

Read More

Items