AI Managing AI (tomtunguz.com)

Recent advancements in AI have led to the emergence of sophisticated models capable of managing other AI systems, with notable examples including Claude, Gemini, and GPT-5. These models can now effectively handle tasks such as code execution and routing requests across various applications, achieving over 90% accuracy in function-calling benchmarks, a significant leap from the sub-50% success rate of earlier models like GPT-4. This improvement highlights the growing importance of large, context-aware models in performing complex management tasks, as smaller models trained solely for specific actions often fail due to a lack of broader world knowledge. The shift toward utilizing "orchestration layers" allows these frontier models to act as executives, coordinating specialized agents that excel in specific domains. This collaborative approach enables organizations to leverage a constellation of AI tools while minimizing failure points and enhancing overall performance. As the industry moves away from monolithic models, new opportunities arise for startups focused on developing specialized agents that integrate seamlessly into these networks. As a result, the future landscape of AI management will likely feature an ecosystem of niche players, redefining how AI technologies interact and drive efficiency.