Kevin Scott argues the industry's focus must shift from scaling laws to application. The reasoning capabilities of AI models have outpaced what's being delivered in products, creating a "capability overhang." The primary challenge now is closing this gap by building more useful agentic systems.
For agents to become truly powerful, they need an open ecosystem similar to the internet. Kevin Scott highlights protocols like MCP and NLweb as foundational layers that serve the same purpose as HTTP and HTML, enabling interoperability and allowing agents to take action across diverse systems.
To combat inefficiency, Kevin Scott is pushing for a standard protocol for all internal Microsoft agents to communicate with its systems. This is a deliberate strategy to fight Conway's Law, which suggests that systems tend to mirror the communication structures of the organizations that build them.
Kevin Scott argues against the belief that open, permissionless systems are inherently less secure than closed ones. He envisions personal security agents using AI to monitor user activity and communications across multiple channels to detect threats, potentially offering more robust security than today's gatekept platforms.
Kevin Scott, a programmer for 41 years and a woodworker, dismisses concerns that agents diminish the craft of coding. He likens the debate to past arguments over power tools versus hand tools in woodworking, stating that agents are simply a new, powerful choice for makers who value different aspects of the process.
Microsoft's CTO observes that the most compelling agent-based startups are not building unique infrastructure. Instead, they are leveraging existing platforms to solve a specific user problem they understand better than anyone else. This signals a market shift from infrastructure-level to application-level innovation.
Kevin Scott predicts that agent interaction will soon shift from a synchronous model (user waits for an immediate response) to an asynchronous one. Users will delegate complex tasks that agents work on over extended periods, iterating and integrating information before reporting back with results.