Over the past three years working in generative AI at Microsoft, I’ve had a front-row seat to some of the most exciting shifts in tech. From chat assistants to multi-modal agents, every layer we’ve built has opened up new questions—and new responsibilities. One of the biggest ones: how do we make the web itself more useful to humans and AI?
Enter NLWeb—a new open project from Microsoft that aims to make websites natively conversational. Instead of keyword searches or rigid menus, users (and agents) can query a site’s content in natural language. For developers and web publishers, NLWeb offers a practical way to tap into your existing structured data—like Schema.org markup or product catalogs—and expose that content intelligently, without reinventing your site.
I’m especially excited that O’Reilly is one of the first contributors. O’Reilly has always been a source of clarity and insight for developers, and now they’re helping imagine how technical content can become even more discoverable, contextual, and actionable in the AI era. They’ve long invested in structured formats for their learning platform—and NLWeb lets them amplify that effort, enabling agents and users to query their vast library like a living, breathing expert system.
This is all made possible by a powerful but accessible architecture and the leadership of R.V. Guha—creator of Schema.org, RSS, and RDF—who recently joined Microsoft as a CVP and Technical Fellow. His vision for a more intelligent, open web is taking root again.
Check out the NLWeb GitHub repo to see what’s possible—and stay tuned as more pioneers like O’Reilly help redefine what it means to “browse” in an AI-native world. Here is a short Q&A to learn more about how O'Reilly is leveraging NLWeb.
Q1: What inspired you to try NLWeb?
We believe the future of AI requires moving beyond monolithic, winner-takes-all approaches. We see NLWeb as a crucial component of an open ecosystem of protocols and tools that will shape the next evolution of the internet—one where conversational interfaces become a primary way of interacting with digital products.
This transformation depends on widespread adoption, which will only happen if the value is distributed among all participants: entrepreneurs, developers, users, and creators alike. By supporting what Tim O'Reilly calls an "architecture of participation for AI," we're helping build a flourishing AI landscape driven by cooperation rather than extraction, creating value for everyone, not just the dominant players.
Q2: How did the setup process go for your team?
Implementation was remarkably straightforward. The beauty of NLWeb is that it leverages the Schema.org metadata we already expose on our website (which AI systems are already crawling). This compatibility allowed us to get a prototype running with our public search in just a few days.
Q3: What’s a query or interaction that made NLWeb really click for you?
Today's content ecosystem spans multiple modalities—live and recorded video, text, audio—covering countless nuanced aspects of technology. This complexity makes traditional search limiting. NLWeb's ability to understand context became clear with queries like "What's the most popular course by Uncle Bob?" where the system recognizes Robert C. Martin's nickname without explicit instruction.
The real breakthrough was seeing the value of multiturn conversations where NLWeb understands evolving intent. An O'Reilly subscriber might start with "I want to build AI apps with foundation models," then refine with "I need books with practical examples" and finally "Show me only books by authors with experience at leading companies." This natural conversation flow delivers increasingly precise results. Keyword search can't do that.
Q4: How are you blending NLWeb with your current experience?
We're working to integrate NLWeb prototypes with our public search functionality. This naturally extends the O'Reilly learning platform's core mission of connecting diverse audiences—from individual practitioners to enterprise teams—with precisely the knowledge they need.
The integration will help transform our search experience from keyword-based discovery to natural conversation, allowing users to articulate complex learning needs in their own words. By implementing this technology early, we're demonstrating to our technically savvy audience the possibilities of conversational interfaces while making our extensive content library more accessible through natural language interaction.
Q5: If NLWeb reaches its full potential, what could it unlock?
Computing history has followed a recurring pattern: Periods of distributed innovation eventually consolidate into monopolistic gatekeepers. An AI ecosystem built on open, standardized protocols like NLWeb could fundamentally disrupt this trajectory.
Just as open source software and the internet fostered experimentation and competition that created broadly distributed value, NLWeb points toward a future where specialized content providers can build AI interfaces directly to their own content. This prevents the extraction model, where large AI companies absorb content to build their own competing services, and instead creates a more distributed ecosystem of value creation and innovation.
Microsoft