The Future of U.S. AI Leadership with Anthropic CEO Dario Amodei (Category AI)
On 11 March 2025, the Council on Foreign Relations held a conversation in its CEO Speaker Series. CFR president Michael Froman interviewed Anthropic’s CEO and co-founder, Dario Amodei. This was the meeting where Dario predicted that AI would write 90% of code within six months and 100% within a year. I decided to watch the whole conversation and note the points I found interesting.
1) Anthropic’s founding and mission Amodei explained why he left OpenAI at the end of 2020 to establish Anthropic. His team, he said, was among the first to recognize the scaling hypothesis: AI improves as compute and data increase. Anthropic was founded as a mission-driven company focused on safety and predictability. Dario clearly implied that OpenAI had different plans for the technology.
2) AI safety and responsible development He described Anthropic’s approaches, including:
- Mechanistic interpretability to understand AI behavior.
- Constitutional AI trained around principles.
- A Responsible Scaling Policy.
- AI safety levels comparable to biosafety levels.
3) Programming’s future This was the hot topic that drew me to the conversation. Amodei predicted that within 3-6 months, AI would be able to write 90% of all code. He also said programmers’ roles would change but remain important for defining requirements, designing applications and making critical decisions.
4) Geopolitics and national security Amodei stressed export controls on AI technologies and discussed risks from possible Chinese dominance. His argument was along the lines that dominance in this technology would bring military and economic superiority more broadly. They also discussed GPU smuggling and cooperation with national-security institutions.
5) Social consequences The discussion addressed AI’s effects on employment, society and human values. Amodei offered optimistic and pessimistic scenarios: AI might help people create great things, or devalue human labor. Interestingly, I am finishing Feeding the Machine, which explores these very issues in an engaging way.
6) Prospects for AI development Amodei discussed possible limits to growth, including potentially exhausting available data by 2030. He suggested that new training approaches, such as reasoning models, could address this by allowing AI to learn from its own thinking.
The conversation closed with reflections on how human qualities and relationships will remain important despite technological progress. AI may help, but will not replace fundamental human values.
#AI #Software #Engineering #Storytelling #ML #Architecture