Alex Albert leads Claude Relations at Anthropic, helping developers turn powerful language models into useful products. As Anthropic’s first prompt engineer and the creator of JailbreakChat, he has worked on both understanding how AI systems fail and designing software that makes their capabilities practical.
Albert grew up in central Washington, studied computer science at the University of Washington, and worked as a software engineer at Google. While still a student, he created JailbreakChat, an open-source collection of prompts that probe chatbot safeguards. He opened the project to contributors in 2023, establishing adversarial model testing as an early focus of his work.
After graduating, Albert joined Anthropic as its first prompt engineer in June 2023. His responsibilities subsequently expanded across developer relations, product, startup partnerships, and developer strategy.
- Prompting should be tested, not mystified. Albert favors test-driven prompt engineering: clear instructions, realistic examples, structured context, and evaluations that reveal whether changes actually improve results.
- Evaluations should reflect practical work. His Claude 3 Opus retrieval experiment showed the model recognizing that an isolated fact looked artificially inserted into unrelated documents, illustrating how contrived benchmarks can distort capability assessments. He also values pull-request evaluations because they require connected reasoning, iteration, and verification.
- AI-native product design demands new interfaces. Albert argues that attaching a chatbot to existing software misses opportunities to rethink the workflow itself. His AI Engineer World’s Fair keynote highlighted Artifacts and Projects: separate workspaces for generated code and documents, combined with shared context from repositories and reference materials.
- Useful agents need tools and durable context. He emphasizes structured outputs, client-side functions, long-context retrieval, and multimodal understanding. He later characterized the Claude Agent SDK as a foundation for general-purpose agents extending beyond coding assistants.
