Kevin Scott on The Future of Programming, AI Agents, and Microsoft’s Big Bet on the Agentic Web
🤖 AI Summary
Overview
This episode explores the transformative impact of AI agents on software engineering, the emergence of the agentic web, and Microsoft's vision for creating open ecosystems that balance innovation and security. Kevin Scott, Microsoft CTO, shares insights from his decades-long career in programming and discusses the future of tools, protocols, and workflows in the age of AI.
Notable Quotes
- Be curious, try stuff. And if it works for you, use it, and if it doesn’t, don’t.
— Kevin Scott, on embracing new tools and technologies.
- You ship your org chart.
— Kevin Scott, explaining the inefficiencies caused by organizational structures in software development.
- Permissionless innovation is the most exciting thing happening right now.
— Kevin Scott, on the value of open ecosystems for creators.
🛠️ The Agentic Web and Its Infrastructure
- Kevin Scott explains that for agents to be truly useful, they must take action on behalf of users by accessing tools, systems, and diverse information sources.
- Microsoft is adopting Model Context Protocol (MCP) as a standard for enabling agents to interact seamlessly across systems, akin to how HTTP standardized the web.
- Emerging protocols like MCP and NLWeb aim to create a shared language for agents, similar to how HTML made websites accessible to browsers.
🔒 Balancing Open Ecosystems and Security
- The debate between open innovation and robust security is a false dichotomy, according to Kevin Scott. AI agents can leverage their reasoning capabilities to enhance security while maintaining openness.
- He envisions agents with identities and entitlement systems that allow them to act on behalf of users securely, querying systems and requesting permissions transparently.
- Open protocols like MCP are key to ensuring security models evolve collaboratively and inclusively.
🧑🎨 The Craftsman’s Perspective on AI Tools
- As a lifelong maker, Kevin Scott reflects on the tension between valuing the craft of creation and optimizing for outcomes with AI tools.
- He compares debates in software engineering to those in woodworking, where purists argue over the use of hand tools versus power tools or CNC machines.
- His advice: embrace curiosity, experiment with new tools, and make choices based on what aligns with your values and goals.
🌐 The Future of Software Engineering Agents
- Kevin Scott predicts a diverse ecosystem of agents tailored to specific problems, driven by startups and developers who deeply understand user needs.
- Developers will continue to enjoy the freedom to experiment with different tools, fostering innovation and variety in agent design.
- He emphasizes that the most impactful agents will focus on solving nuanced problems rather than reinventing infrastructure.
⏳ From Synchronous to Asynchronous Agentic Workflows
- The current mode of interacting with agents—issuing prompts and waiting for immediate responses—will evolve into asynchronous workflows.
- As agents gain better reasoning and planning capabilities, users will delegate complex tasks and let agents work independently over extended periods.
- This shift will enable agents to tackle ambitious, multi-step problems, integrating diverse inputs and iterating until ready for user action.
AI-generated content may not be accurate or complete and should not be relied upon as a sole source of truth.
📋 Episode Description
I interviewed Microsoft CTO Kevin Scott about the future of agents and software engineering for another special edition of AI & I.
With 41 years of programming behind him, Kevin has lived through nearly every big shift in modern software development. Here’s his clear-eyed take on what’s changing with AI, and how we can navigate what’s next:
The real breakthrough for the agentic web is better plumbing. Kevin thinks agents won’t be useful until they can take action on your behalf by using tools and fetching data. To do this, agents need access across your systems—and Microsoft’s answer is adopting Model Context Protocol, or “MCP,” that allows an agent to access tools and fresh data beyond its knowledge base, as their standard protocol for agents to move through contexts and get things done.
How the agentic web echoes the early internet. Just as protocols like HTTP and HTML gave the web a shared language, Kevin believes the agentic web needs its own infrastructure—the first glimpses of this include MCP (the HTTP of agents) and NLWeb, Microsoft’s push to make websites legible to agents (similar to what HTML did for browsers).
Open ecosystems can coexist with strong security systems. Kevin argues that the “tradeoff” between ecosystems that allow “permissionless” innovation and robust security is a false dichotomy. With AI agents that understand your personal risk preferences—and know your communication habits across email, text, and other channels—they could detect when something suspicious is happening and act on your behalf.
The craftsman’s dilemma in the age of agents. Kevin is a lifelong maker—of software, ceramics, even handmade bags—and he cares deeply about how things are made. Because this can feel at odds with coding with AI agents, Kevin’s approach is to notice where the process matters most to him, and where it's okay to optimize for outcomes. After four decades of seeing breakthrough technologies, his advice is simple: be curious, try stuff, and use it if it works for you.
The future of software engineering agents is plural. Kevin believes the future of software engineering agents will be diverse because developers who enjoy the freedom of playing with different tools is one of the most consistent patterns he’s seen in his decades in tech. What will drive this diversity, he says, is builders who deeply understand specific problems and tailor agents to solve them exceptionally well.
How agentic workflows will evolve. Kevin sees a shift from short back-and-forth interactions with agents to longer, async feedback loops. As the agentic web matures and model reasoning improves, people will start handing off bigger, more ambitious tasks and letting agents run with them.
Timestamps: