I’m actually wanting ahead to our second O’Reilly AI Codecon, Coding for the Agentic World, which is occurring on September 9, on-line from 8am to midday Pacific time, with a follow-on day of further demos on September 16. However I’m additionally wanting ahead to how the AI market itself unfolds: the stunning twists and turns forward as customers and builders apply AI to real-world issues.
The pages linked above give particulars on this system for the occasions. What I need to give here’s a little bit of the why behind this system, with a bit extra element on a few of the fireplace chats I will likely be main.
From Invention to Software
There was a lot focus up to now on the large AI labs, the mannequin builders, and their razzle-dazzle about AGI, and even ASI. That narrative implied that we had been heading towards one thing unprecedented. But when it is a “regular expertise” (albeit one as transformational as electrical energy, the interior combustion engine, or the web), we all know that LLMs themselves are only the start of a protracted means of discovery, product invention, enterprise adoption, and societal adaptation.
That means of collaborative discovery of the actual makes use of for AI and reinvention of the companies that use it’s occurring most clearly within the software program trade. It’s the place AI is being pushed to the bounds, the place new merchandise past the chatbot are being launched, the place new workflows are being developed, and the place we perceive what works and what doesn’t.
This work is commonly being pushed ahead by people, who’re “studying by doing.” A few of these people work for giant corporations, others for startups, others for enterprises, and others as impartial hackers.
Our focus in these AI Codecon occasions is to clean adoption of AI by serving to our prospects reduce by means of the hype and perceive what’s working. O’Reilly’s mission has at all times been altering the world by sharing the data of innovators. In our occasions, we at all times search for people who find themselves on the forefront of invention. As outlined within the name to motion for the primary occasion, I used to be involved concerning the chatter that AI would make builders out of date. I argued as a substitute that it might profoundly change the method of software program improvement and the roles that builders do, however that it might make them extra necessary than ever.
It seems like I used to be proper. There’s a big ferment, with a lot new to be taught and try this it’s a extremely thrilling time to be a software program developer. I’m actually excited concerning the practicality of the dialog. We’re not simply speaking concerning the “what if.” We’re seeing new AI powered companies assembly actual enterprise wants. We’re witnessing the shift from human-centric workflows to agent-centric workflows, and it’s occurring quicker than you suppose.
We’re additionally seeing widespread adoption of the protocols that can energy all of it. When you’ve adopted my work from open supply to Net 2.0 to the current, you recognize that I imagine strongly that essentially the most dynamic programs have “an structure of participation.” That’s, they aren’t monolithic. The boundaries to entry should be low and enterprise fashions fluid (no less than within the early phases) for innovation to flourish.
When AI was framed as a race for superintelligence, there was a robust expectation that it might be winner takes all. The primary firm to get to ASI (and even simply to AGI) would quickly be thus far forward that it might inevitably grow to be a dominant monopoly. Builders would all use its APIs, making it into the only dominant platform for AI improvement.
Protocols like MCP and A2A are as a substitute enabling a decentralized AI future. The explosion of entrepreneurial exercise round agentic AI jogs my memory of the most effective type of open innovation, very like I noticed within the early days of the non-public laptop and the web.
I used to be going to make use of my opening remarks to sound that theme, after which I learn Alex Komoroske’s marvelous essay, “Why Centralized AI Is Not Our Inevitable Future.” So I requested him to do it as a substitute. He’s going to present an up to date, developer-focused model of that as our kickoff discuss.
Then we’re going into a piece on agentic interfaces. We’ve lived for many years with the GUI (both on computer systems or cellular purposes) and the online because the dominant methods we use computer systems. AI is altering all that.
It’s not simply agentic interfaces, although. It’s actually growing true AI-native merchandise, seeking out the chances of this new computing cloth.
The Nice Interface Rethink
Within the “regular expertise” framing, a elementary expertise innovation is distinct from merchandise based mostly on it. Consider the invention of the LLM itself as electrical energy, and ChatGPT because the equal of Edison’s incandescent mild bulb and the event of the distribution community to energy it.
There’s a little bit of a lesson in the truth that the telegraph was the primary large-scale sensible software of electrical energy, over 40 years earlier than Edison’s lightbulb. The phone was one other killer app that used electrical energy to energy it. However regardless of their scale, these had been specialised gadgets. It was the infrastructure for incandescent lighting that turned electrical energy right into a general-purpose expertise.
The world quickly noticed electrical resistance merchandise like irons and toasters, and electrical motors powering not simply factories however family home equipment similar to washing machines and ultimately fridges and air con. Many of those family merchandise had been plugged into mild sockets, because the pronged plug as we all know it at the moment wasn’t launched till 30 years after the primary mild bulb.
The lesson is that in some unspecified time in the future within the improvement of a basic objective expertise, product innovation takes over from pure expertise innovation. That’s the part we’re getting into now.
Have a look at the evolution of LLM-based merchandise: GitHub Copilot embedded AI into Visible Studio Code; the interface was an extension to VS Code, a 10-year-old GUI-based program. Google’s AI efforts had been tied into its web-based search merchandise. ChatGPT broke the mildew and launched the primary radically new interface because the net browser. All of the sudden, chat was the popular new interface for every thing. However Claude took issues additional with Artifacts after which Claude Code, and as soon as coding assistants gained extra advanced interfaces, that kicked off at the moment’s fierce competitors between coding instruments. The subsequent revolution is the development of a brand new computing paradigm the place software program consists of clever, autonomous brokers.
I’m actually wanting ahead to Rachel-Lee Nabors’s discuss on how, with an agentic interface, we’d transcend the normal browser: AI brokers can adapt content material on to customers, providing privateness, accessibility, and suppleness that legacy net interfaces can not match.
However it appears to me that there will likely be two sorts of brokers, which I name “demand aspect” and “provide aspect” brokers. What’s a “demand aspect” agent? As a substitute of navigating advanced apps, you’ll merely state your aim. The agent will perceive the context, entry the mandatory instruments, and current you with the end result. The imaginative and prescient continues to be science fiction. The truth is commonly a kludge powered by browser use or API calls, with MCP servers more and more providing an AI-friendlier interface for these demand-side brokers to work together with. However why ought to it cease there? MCP servers are static interfaces. What if there have been brokers on either side of the dialog, in a dynamic negotiation? I believe that whereas demand-side brokers will likely be developed by enterprise funded startups, most server-side brokers will likely be developed by enterprises as a type of conversational interface for each people and AI brokers that need entry to their advanced workflows, knowledge, and enterprise fashions. And people enterprises will typically be utilizing agentic platforms tailor-made for his or her use. That’s a part of the “provide aspect agent” imaginative and prescient of corporations like Sierra. I’ll be speaking with Sierra cofounder Clay Bavor about this subsequent step in agentic improvement.
We’ve grown accustomed to fascinated with brokers as lonely customers—“inform me the climate,” “scan my code,” “summarize my inbox.” However that’s solely half the story. If we construct supply-side agent infrastructure—autonomous, discoverable, ruled, negotiated—we unlock agility, resilience, safety, and collaboration.
My curiosity in product innovation, not simply advances within the underlying expertise, can be why I’m enthusiastic about my fireplace chat with Josh Woodward, who co-led the staff that developed NotebookLM at Google. I’m an enormous fan of NotebookLM, which in some ways introduced the facility of RAG (retrieval-augmented technology) to finish customers, permitting them to gather a set of paperwork right into a Google drive, after which use that assortment to drive chat, audio overviews of paperwork, research guides, thoughts maps, and far more.
NotebookLM can be a beautiful solution to construct on the deep collaborative infrastructure offered by Google Drive. We have to suppose extra deeply about collaborative interfaces for AI. Proper now, AI interplay is generally a solitary sport. You’ll be able to share the outputs with others, however not the generative course of. I wrote about this not too long ago in “Individuals Work in Groups, AI Assistants in Silos.” I feel that’s an enormous miss, and I’m hoping to probe Josh about Google’s plans on this space, and desperate to see different improvements in AI-mediated human collaboration.
GitHub is one other present software for collaboration that has grow to be central to the AI ecosystem. I’m actually wanting ahead to speaking with outgoing CEO Thomas Dohmke each concerning the ways in which GitHub already gives a type of exoskeleton for collaboration when utilizing AI code-generation instruments. It appears to me that one of many frontiers of AI-human interfaces will likely be people who allow not simply small groups however ultimately massive teams to collaborate. I believe that GitHub could have extra to show us about that future than we now suspect.
And eventually, we are actually studying that managing context is a important a part of designing efficient AI purposes. My cochair Addy Osmani will likely be speaking concerning the emergence of context engineering as an actual self-discipline, and its relevance to agentic AI improvement.
Instrument-Chaining Brokers and Actual Workflows
At the moment’s AI instruments are largely solo performers—a Copilot suggesting code or a ChatGPT answering a question. The subsequent leap is from single brokers to interconnected programs. This system is stuffed with classes on “tool-to-tool workflows” and multi-agent programs.
Ken Kousen will showcase the brand new technology of coding brokers, together with Claude Code, Codex CLI, Gemini CLI, and Junie, that assist builders navigate codebases, automate duties, and even refactor intelligently. In her discuss, Angie Jones takes it additional: brokers that transcend code technology to handle PRs, write exams, and replace documentation—stepping “out of the IDE” and into real-world workflows.
Much more thrilling is the concept of brokers collaborating with one another. The Demo Day will showcase a multi-agent coding system the place brokers share, right, and evolve code collectively. This isn’t science fiction; Amit Rustagi’s discuss on decentralized AI agent infrastructure utilizing applied sciences like WebAssembly and IPFS gives a sensible architectural framework for making these agent swarms a actuality.
The Essential Ingredient: Widespread Protocols
How do all these brokers discuss to one another? How do they uncover new instruments and use them safely? The reply that echoes all through the agenda is the Mannequin Context Protocol (MCP).
A lot because the distribution community for electrical energy was the enabler for all the product innovation of {the electrical} revolution, MCP is the foundational plumbing, the common language that can permit this new ecosystem to flourish. A number of classes and a whole Demo Day are devoted to it. We’ll see how Google is utilizing it for agent-to-agent communication, how it may be used to manage advanced software program like Blender with pure language, and even the way it can energy novel SaaS product demos.
The heavy concentrate on a standardized protocol alerts that the trade is maturing previous cool demos and is now constructing the strong, interoperable infrastructure wanted for a real agentic economic system.
If the event of the web is any information, although, MCP is a starting, not the top. TCP/IP turned the muse of a layered protocol stack. It’s seemingly that MCP will likely be adopted by many extra specialised protocols.
Why This Issues
| Theme | Why It’s Thrilling |
|---|---|
| Autonomous, Distributed AI | Brokers that chain duties and function behind the scenes can unlock totally new methods of constructing software program. |
| Human Empowerment & Privateness | The push in opposition to centralized AI programs is a reminder that instruments ought to serve customers, not management them. |
| Context as Structure | Elevating enter design to first-class engineering—this can vastly enhance reliability, belief, and AI habits over time. |
| New Developer Roles | We’re seeing builders transition from writing code to orchestrating brokers, designing workflows, and managing programs. |
| MCP & Community Results | The concept of an “AI-native net,” the place brokers use standardized protocols to speak, is highly effective, open-ended, and stuffed with alternative. |
I look ahead to seeing you there!
We hope you’ll be part of us at AI Codecon: Coding for the Agentic World on September 9 to discover the instruments, workflows, and architectures defining the subsequent period of programming. It’s free to attend. Register now to avoid wasting your seat. And be part of us for O’Reilly Demo Day on September 16 to see how consultants are shaping AI programs to work for them by way of MCP.