Anthropic Launches New Prompt Engineering Tools in Developer Console

Posted on November 14, 2024

Anthropic has unveiled new prompt improvement and example management features in their Console to help developers create more effective AI applications. The new prompt improver automatically refines existing prompts using techniques like chain-of-thought reasoning, example standardization, and prefill addition. Testing shows these improvements increased accuracy by 30% for multilabel classification and achieved 100% word count adherence in summarization tasks.

The update also introduces structured example management, allowing developers to add and edit input/output pairs directly in the Workbench. Users can leverage Claude to generate synthetic examples automatically. A new prompt evaluator includes an 'ideal output' column for grading model outputs on a 5-point scale.

Technology company Kapa.ai has already implemented these tools to migrate their AI workflows to Claude 3.5 Sonnet. According to Co-Founder Finn Bauer, the prompt improver helped them achieve faster production deployment. These features are now accessible to all Anthropic Console users.