The Agentic Web: My Journey to Finding the Future of the Web in an Agentic World
The UI Revolution: How Conversational AI Services will Render Native UI/UX and Why Browser Agents are the Key to Unlocking New Customer Workflows.
The Next Decade of Web Profitability: Why Your Business Model is Under Threat from UI-Driven AI
As a front-end developer, my main concern is fundamentally around user experience—and, by extension, the browser experience. There has been a wealth of conversation around AI and agents, but most of it has been focused on back-end infrastructure, often utilizing Python and libraries such as Langchain. For a while, it was difficult for me to determine my role in this coming wave of new technology.
The Unseen Shift in Customer Interaction
About two years ago, I had a flawed premise. I initially thought AI would surely be able to complete forms because it possesses powerful image recognition and strong language understanding, and forms are fundamentally built upon HTML and JavaScript. That idea quickly faded amidst the broader industry noise.
However, the idea resurfaced as agents began operating on the internet to either complete complex tasks (such as acquisitions) or conduct general research (like reading and synthesizing information). The way we structure our web applications is going to need to change.
Of course, at that point, I didn't truly grasp the depth of Agentic AI. I knew that MCP (Multimodal Communicative Protocol) was emerging and would be significant. I understood that agents had the potential to be valuable and capable of some pretty amazing things.
The New Interface: UI Inside the Conversation
However, I couldn't get a clear picture of how all of this would connect. I had another insight: surely, with AI assistants and conversational services like Claude or Gemini showing text responses to prompts, there had to be a world where we could deliver UI back to the conversational AI.
For instance, if a customer is trying to choose highly customized options when buying your product, trying to fine-grain those selections purely through text-based conversational AI wouldn't make sense. You would need a rich user interface, and that UI would represent the optimal solution for user experience.
Therefore, it became apparent to me that one day we would see HTML, JavaScript, and CSS UI components rendered within conversational AIs. Part of that journey would involve a prompt or voice interaction, and the other part would be a conventional mouse-click visual interaction. This is where efficiency and conversion rates will live.
The Rise of Autonomous Agents and the CPA Model
On the flip side, we are beginning to see agents directly operating in the browser. Atlas from OpenAI now features agentic web browsing capabilities, and if you look at one of my LinkedIn posts, I demonstrate how I actually created this blog platform using agents through the browser. We're seeing similar developments from startups like Perplexity and Comet's browser.
Naturally, the significant debate, voiced by the CEO of Perplexity, is that Google is under threat because the CPC (Cost-Per-Click) model they’ve profited on for decades is being challenged by a CPA (Cost-Per-Acquisition) model. I will, however, be writing more blog posts about that in the future.
The Value of Agentic Workflows
The idea of having an agent in the browser capable of performing complex UI workflows is certainly something that will be highly valuable to consumers. I sometimes laugh when people say AI will take away our humanity. I think to myself, "Is being human really defined by attempting to accomplish a task through a cumbersome mobile or desktop browsing workflow?" Please, if anyone has a suitable answer, let me know.
From my day-to-day experience—trying to scroll, click, and navigate through multiple tabs and UIs to book a flight, synchronize my train journey to the airport, check-in time, check-out time, and coordinate bag storage at a hotel with a late check-out—it is simply a headache. I don't see anything inherently human in that at all.
So, it is quite clear that if we can correctly implement the ReAct prompting side of Agency, we are going to see significant benefits in our daily lives and workflows. For your business, this translates directly to removing friction from the path to purchase.
The Core Protocols Driving the Change
What are the technologies driving this shift? It appears that MCP is more than likely going to be the central protocol. While early criticisms of the protocol are beginning to surface (which I’ll be writing about later), I place my bets on MCP being the technology that fuels these experiences.
From my initial, deeper, and intent-filled research into MCP and the web, there seem to be two projects that deserve very careful consideration as the potential future of the web and artificial intelligence:
- WebMCP: Driven by the W3C community working group.
- MCP-UI: Which I believe is driven by Shopify.
Each project addresses the challenge of agency, whether it involves an agent operating within the browser or UI being sent back to assistive AI services.
I will be researching these projects in great detail with the aspiration of making them clear and understandable to a non-technical audience, as well as providing learning resources and evangelism around these technologies should they become standardized and important industry tools.