In the interview, Anthropic’s Chief Product Officer Mike Krieger discusses the advancements in their next-generation AI model, Claude 4. This new version represents a significant upgrade, capable of working for extended periods and handling hours of complex tasks. Krieger highlights that Claude 4 can manage its own memory and operate autonomously across various hard tasks, including coding and research, making it a more effective virtual collaborator rather than just a simple assistant. Early access partners have already begun deploying Claude 4 for diverse use cases, demonstrating its improved capabilities and longer horizon work.
Krieger emphasizes that Claude 4 sets a new standard in human-AI collaboration by shifting the relationship from single questions and answers to a more dynamic, multi-tasking, and parallel approach. He describes how teams are now able to delegate multiple tasks simultaneously and manage AI systems more like orchestrators or managers, rather than just users engaging in one-on-one conversations. This evolution allows for more sophisticated workflows and enhances productivity, positioning Claude 4 as a true virtual collaborator.
Addressing concerns about hallucinations—errors where AI fabricates information—Krieger explains that Anthropic has implemented a citations API to ground AI responses in real sources. When Claude references external information, it includes direct links to the sources, whether web pages or documents stored in drives. This approach significantly reduces hallucinations by ensuring that all information provided is verifiable and traceable, improving the reliability of AI outputs in research and workflow applications.
Krieger compares Anthropic’s focus on “agentic behavior” with other major AI players, noting that their models excel in acting across tools, systems, and time horizons. While companies like OpenAI focus on multi-modality (voice, images) and Google emphasizes low latency and smaller models, Anthropic prioritizes models capable of long-term, multi-task, and cross-system work. This focus on agentic capabilities positions Claude as a leader in performing complex, knowledge-intensive tasks that require sustained, autonomous effort.
Finally, Krieger discusses the company’s approach to hardware and partnerships, particularly with Amazon, to enhance AI integration into everyday devices. He notes that Anthropic does not currently develop hardware but collaborates with companies like Amazon to embed AI into existing devices, such as Alexa. He also reflects on his first year at Anthropic, comparing it to his experience at Instagram, emphasizing the rapid pace of AI development and the importance of staying on the frontier. Additionally, he mentions evolving hiring practices to allow candidates to use AI tools during interviews, recognizing their integral role in modern software development.
