AI Codecon: Coding for the Agentic World

Schedule

The time frames are only estimates and may vary according to how the class is progressing.

Introduction – Tim O’Reilly (10 minutes)

• Tim welcomes you to Coding for the Agentic World.

Part 1: Agentic Interfaces

Why Centralized AI Is Not Our Inevitable Future – Alex Komoroske (15 minutes)

• Tim O’Reilly has been writing and speaking about the need for “an architecture of participation” for AI, and how the endgame is not a single dominant platform. Alex Komoroske, CEO of Common Tools, proposes a path to that future, which includes data sovereignty, open ecosystems, and personalized AI that works exclusively for the user. He’ll explore the structural dangers of a centralized AI future and present a compelling case for a distributed, pluralistic AI ecosystem where human agency and privacy are paramount. His goal is to encourage developers to build tools that empower individuals rather than a few large companies.

Josh Woodward and Tim O’Reilly Fireside Chat (20 minutes)

• NotebookLM was one of the first breakout successes beyond simple chatbots and coding tools. Besides the originality of some of the things it does, it’s a great illustration of how to rethink existing apps and services to give them a truly useful AI native interface. Josh Woodward, who leads Google Labs and the Gemini app, shares the origin story of NotebookLM and the lessons developers might take from it. And of course, Josh and Tim are going to talk about what’s next.

Context Engineering – Addy Osmani (15 minutes)

• In Andrej Karpathy’s words, context engineering is “one small piece of an emerging thick layer of non-trivial software” that powers real LLM apps. It’s the evolution of prompt engineering, reflecting a broader, more system-level approach. Addy Osmani, engineering leader at Google, explores this practical, systems-based approach to structuring all the information an LLM needs to perform reliably and usefully. You’ll hear how you can dynamically assemble task-specific context using data, instructions, tools, and memory, and how good input design translates to better, more trustworthy output.

Death of the Browser – Rachel-Lee Nabors (15 minutes)

• The internet has turned into a maze of walled gardens, ad-powered content, and algorithmic manipulation. AI agents are the key to putting users back in the driver’s seat. React team alum and web standards advocate Rachel-Lee Nabors explores how agents can create interfaces that adapt to a user’s needs and preferences while providing privacy and accessibility features that traditional browsers can’t match. Rachel-Lee also discusses the technical implications of this shift and explores emerging and existing technologies and methods for direct content distribution and access.

Break (5 minutes)

Part 2: Tool-to-Tool Workflows

Clay Bavor and Tim O’Reilly Fireside Chat (20 minutes)

• Join Tim O’Reilly and Clay Bavor, cofounder of Sierra, for a forward-looking conversation that challenges conventional thinking about AI agents. While much of the current discourse focuses on agents as individual productivity tools, Bavor presents a compelling vision where AI agents become the primary digital interface for enterprises—as essential as websites and mobile apps—and represent the true future of artificial intelligence in enterprise environments. The discussion explores Sierra’s unique approach, including its platform for creating singular branded agents that power entire customer experiences and its business model, which ensures companies only pay when these agents deliver results.

The Hitchhiker’s Guide to AI Coding Agents (Don’t Panic!) – Ken Kousen (15 minutes)

• In the occasionally bewildering universe of AI-powered coding tools, a new wave of agentic assistants is charting fresh territory for developers. From Claude Code and Codex CLI to Junie and Gemini CLI, these agents promise to navigate codebases, automate tasks, and even refactor your work. Ken Kousen, author of several O’Reilly books and courses, introduces the key players, shows quick demos of what they can (and can’t) do, and shares tips on making them work for you, not the other way around.

Beyond Code Generation: Getting Real Work Done with AI Agents – Angie Jones (15 minutes)

• AI tools can write code now. That’s cool. But what about everything else? Triaging issues, writing tests, opening PRs, updating docs: This is the stuff we should be delegating to AI. Angie Jones, global VP of developer relations at Block Inc., shows you what happens when agents step out of the IDE and actually start doing work. No flashy prompts—just practical, repeatable workflows where agents take real action across environments and tools, and collaborate with other agents and models to get complex tasks done faster. If you’re using AI to help you write code, great. Now let it help you ship.

Your AI Agents Don’t Need You Anymore: Building Self-Organizing, Decentralized AI Agent Infrastructure – Amit Rustagi (5 minutes)

• AI and big data technologist Amit Rustagi presents a cutting-edge architectural framework for decentralized, self-organizing AI agent systems, leveraging WebAssembly (Wasm) and IPFS to enable adaptive autonomy beyond traditional orchestration. By combining Wasm’s portable, sandboxed execution with IPFS’s decentralized storage and content addressing, the framework facilitates peer-to-peer agent coordination, evolutionary specialization, and trustless collaboration without centralized oversight. Amit also critically examines the trade-offs: Wasm’s limited hardware acceleration, IPFS latency in real-time scenarios, and the ethical implications of fully autonomous credit economies. Designed for the postcloud era, this framework offers a modular, evolution-ready infrastructure—combining the latest in decentralized compute, storage, and cryptography—while prompting urgent dialogue on safety, accountability, and the limits of emergent machine autonomy in production systems.

Part 3: Background Coding Agents

Kathy Korevec and Addy Osmani Fireside Chat (20 minutes)

• Join Kathy Korevec and Addy Osmani for an inside look at how the AIDA team at Google Labs is building at the intersection of now and next. They’ll explore the making of Jules—an AI-powered developer tool designed not just to add intelligence, but to remove friction and unlock entirely new creative workflows. Learn how the team balances rapid AI evolution with thoughtful, high-integrity product design, crafting tools that anticipate developer needs without getting in the way. It’s a messy, nonlinear process—but one rooted in vision, precision, and a deep respect for developers.

Break (5 minutes)

Agentic Web: MCP, A2A, MIT Project NANDA, and Beyond – Ramesh Raskar (10 minutes)

• The Model Context Protocol (MCP), Agent2Agent (A2A), NLWeb, and MIT’s Project Nanda have sparked a new wave of excitement around the agentic web, setting the stage for rapid innovation and collaboration. Open protocols like MCP and A2A are enabling AI agents to interoperate seamlessly, making it easier for developers and enterprises to build powerful, agent-driven applications. But what are the opportunities in the next phases? How will agentic commerce evolve? What will be the emergent intelligence in agentic societies? Ramesh Raskar, an associate professor at MIT Media Lab, considers these questions in relation to Project Nanda, which is architecting the building blocks for the internet of AI agents in three phases—foundations, commerce, and societies.

The Team of Tomorrow: Agents, Alignment, and Team Ritual – Chris Butler (5 minutes)

• Chris Butler, a product operations leader in AI and productivity at GitHub, imagines a “kickoff meeting” in the near future where engineers, designers, product managers, and AI agents come together to start a new project. Rather than replacing team members, the agents in the room provoke, question, and draft alongside them and actively shape the work, acting as representatives of customer behavior, technical constraints, and ethical risks. Through this design fiction, Chris explores how today’s teams can learn to adapt rituals, reimagine roles, and stay grounded in purpose while building with intelligent collaborators.

From Proof to Platform: Scaling Agentic AI in a Fortune 300 Software Enterprise – Brennon Bortz (5 minutes)

• As agentic AI systems mature, the challenge is no longer proving they can work, but proving they can scale. Leidos is working to embed agentic capabilities into the day-to-day flow of software development across a global workforce of thousands. Brennon Bortz, chief engineer for corporate R&D at Leidos, describes how the company identified high-leverage opportunities for agentic automation within the SDLC, structured pilots to compare agentic and nonagentic performance, integrated tools into secure, developer-friendly platforms, managed adoption across a risk-sensitive enterprise, and defined new success metrics and evolving roles to support long-term change.

Part 4: MCP and Agent Protocols

Designing for AI Agents: MCP – Jessica Kerr (15 minutes)

• These days, software isn’t easy for people to use until it’s easy for their AI agents to use. That means we need a great MCP. Model Context Protocol servers give AI agents fingers to feel around and act in the world, one tool at a time. Jessica Kerr, engineering manager of developer relations at Honeycomb, shares design considerations from the Honeycomb MCP, contrasts MCP design with human UX and software APIs, and discusses ways to tell whether the MCP is effective in production.

Blender MCP – Siddharth Ahuja (15 minutes)

• Blender MCP is a tool that allows AI language models like Claude to control Blender through the Model Context Protocol, reducing the complexity of 3D modeling from hours to minutes by letting users create scenes, assets, and animations through natural language prompts. The project gained significant traction with 12.8K GitHub stars and over 280K downloads. Siddharth Ahuja, creator of Blender MCP and Ableton MCP, explains how Blender MCP was made and how MCPs for creative tools enable anyone to be a creator. Examples include generating terrain, creating animated scenes, and even combining 3D modeling with music creation through an Ableton MCP integration.

Thomas Dohmke and Tim O’Reilly Fireside Chat (20 minutes)

• Tim O'Reilly speaks with Thomas Dohmke, GitHub’s outgoing CEO, about the company’s role as the backbone of collaborative coding and how AI agents are becoming integral team members in the software development lifecycle. They’ll discuss how GitHub has become essential infrastructure for AI coding—with agents operating similarly to traditional CI/CD processes but creating new challenges around security and repository management. They’ll also explore how the collapse of traditional boundaries between product management, design, and engineering enabled by AI is transforming software development roles into “full stack builders” who can take ideas from conception to deployment.

Closing Remarks – Tim O’Reilly and Addy Osmani (5 minutes)

• Tim and Addy close out today’s event.