Microsoft VP Ray Smith on AI’s Future

The impact of DeepSeek’s R1 model on the current AI field is significant. The emergence of this model shows that it is possible to achieve high-level thinking capabilities with significantly lower resource investments. This development has sparked innovation in the AI ecosystem, pushing the boundaries of what AI can achieve.

Microsoft’s Vice President of AI Agents views this development as an opportunity for the industry to accelerate progress, benefit end-users, and continue pushing the boundaries of what AI can achieve. The company supports a choice of models and uses the right models for the right tools.

The cost of inference is decreasing, and applications are already available. The industry is moving towards a future where AI agents will play a significant role in various sectors. AI agents are tools with the ability to perform a thinking process, plan, and evaluate their input to decide what actions to take.

These agents are already being used in businesses, such as a British pet retailer using AI agents connected to their customer relationship management system to handle customer inquiries. A chemical company has built an AI agent that optimizes its supply chain and identifies potential issues.

The consumer sector is expected to see the introduction of AI agents that can access personal information and perform actions like planning entertainment. For example, AI agents may soon be able to suggest and book a restaurant based on a user’s preferences.

In the near future, AI agents will be available on smartphones, interacting with users and performing tasks. The necessary capabilities, such as personalization and real-time voice interaction, already exist.

Within six months, AI agents are expected to break into existing consumer apps and industries, providing a competitive edge to companies. However, there are tasks that AI models may still not be able to replace, even in 10 years, such as negotiation, strategy, and tasks requiring emotional intelligence.

Human programmers will still be needed, as AI models can generate apps or software more independently, but human skills are required to make it all work together. AI models will follow the path of becoming more efficient and cost-effective, just like personal computers did.

The ultimate goal is to democratize AI and help people everywhere use these tools to develop their skills. As model costs decrease and become more accessible, there will be greater usage and widespread adoption of AI agents.