Anthropic Launches Auto-Tuning Workflow for Claude Code

iconKuCoinFlash
Share
AI summary iconSummary
Anthropic announced a new auto-tuning workflow for the Claude API skill of Claude Code, according to real-world assets (RWA) news. The system constructs test sets from real conversations, bugs, or manual cases, then iteratively adjusts prompts, model settings, and reasoning depth. Successful changes are retained; failures are rolled back. Generalization is evaluated using reserved test cases. In a demo involving 44 customer service tickets, accuracy on new cases increased from 78.6% to 90.5%, with an 80% reduction in cost. The update aligns with broader cryptocurrency trends in AI-driven efficiency.
ME AI message: Anthropic has added an automated tuning workflow to the official claude-api Skill for Claude Code. The build-eval phase first creates a test set and scoring method from real conversations, bugs, or manual cases; hillclimb then has Claude Code repeatedly modify the system prompt, tool descriptions, model, and reasoning intensity. After each change, the tests are rerun—improvements are kept, and regressions are rolled back. To prevent Claude from over-optimizing for the test cases, not all samples are used together for tuning. Some cases are used to identify issues and adjust configurations, while another set is reserved to check whether the changes generalize to unseen problems. If scores improve on the tuning set but show no improvement on the reserved test set, the modification may be flagged as overfitting and reverted. Anthropic demonstrated this workflow using 44 customer support tickets. Claude Code sequentially cleaned the prompt, added business rules, experimented with different models, and reduced reasoning intensity. Ultimately, it found an optimal configuration using Sonnet 5 with low effort. On 14 tickets not involved in tuning, accuracy improved from 78.6% with the initial configuration to 90.5%, while costs dropped to about one-fifth of the original. (Source: BlockBeats)
Source:Show original
Disclaimer: The information on this page may have been obtained from third parties and does not necessarily reflect the views or opinions of KuCoin. This content is provided for general informational purposes only, without any representation or warranty of any kind, nor shall it be construed as financial or investment advice. KuCoin shall not be liable for any errors or omissions, or for any outcomes resulting from the use of this information. Investments in digital assets can be risky. Please carefully evaluate the risks of a product and your risk tolerance based on your own financial circumstances. For more information, please refer to our Terms of Use and Risk Disclosure.