Are conversations the future of UX?

This morning, I asked ChatGPT to retrieve the latest working draft of an article from my blog.

We talked through where the ideas could be richer. I asked it to draft additions, then save them directly into Statamic. After reading through the result, I asked it to tighten the writing, remove repetition, and keep each section focused.

So it did. All in the same conversation, working with the actual content in my CMS.

And yeah, that's pretty damn cool. 😄

But the more I thought about it, the more I realized that the exciting part wasn't just editing my blog through ChatGPT. It was how little I needed to interact with the CMS itself.

And that raised a much bigger question about the future of how we interact with software.

Okay, what is MCP?

MCP (Model Context Protocol) is an open standard for connecting AI applications to external tools and data. I recently took Anthropic's course on MCP to understand how the protocol works under the hood, so connecting it to my own site has been a pretty satisfying way to put that learning into practice.

In my case, an MCP integration gives ChatGPT access to tools for retrieving articles, inspecting Statamic's content structure, and making changes. MCP doesn't magically make the CMS understand natural language; the integration exposes specific capabilities, with permissions and structured operations behind them.

As a developer, that's almost as interesting as the experience itself. The conversation feels natural because there's a well-defined technical interface doing the work underneath.

Less clicking, more thinking

The usual AI-assisted writing workflow involves a surprising amount of admin, and while none of it is particularly difficult, it interrupts the part I actually enjoy: figuring out what I want to say.

Now I can retrieve a draft, discuss changes, compare it with earlier articles, and save an updated version without constantly switching between tools. I can also prepare preview images, related articles, metadata, and outside perspectives in the same workflow.

The context matters too. ChatGPT can read my existing posts and work with the article that's actually in Statamic, rather than relying on whatever I remember to paste into a prompt. It still needs direction, and it can absolutely produce a sentence I'd never say, but it's a much more useful editorial conversation when we're both looking at the same material.

From managing content to managing intent

Platforms like WordPress and Statamic made publishing accessible to people who don't need to understand the underlying code. Their interfaces abstract away technical complexity. But what happens when natural language becomes another layer of abstraction?

But what happens when natural language becomes an even more accessible abstraction?

Traditionally, the CMS dictates the workflow. We learn where the menus are, which fields to edit, how to upload assets, and which buttons to click. Now I can simply describe the outcome I want: "Find my latest article, expand this idea, remove repetition, and save it as a draft."

The assistant coordinates the operations while I focus on the result. Instead of learning every step of a workflow, I can describe what I want to accomplish. That's the shift I find interesting.

What does this mean for traditional UX?

We've spent decades making software easier to navigate: clearer menus, fewer clicks, better forms. AI agents challenge a deeper assumption: that users must navigate an application's interface to achieve their goals.

Imagine asking a CMS to review an entire content library for outdated information, identify inconsistencies, and prepare coordinated changes for approval. Or asking a CRM to summarize neglected opportunities and draft follow-ups. Those are tasks that can span dozens of screens and several different workflows.

That doesn't mean chat is always better. Visual editing, precise controls, comparisons, and review screens still matter. Sometimes a button is faster than explaining what you want, and agents can misunderstand instructions. The goal isn't to eliminate every interface — it's to eliminate unnecessary steps.

I think the best UX will increasingly be judged not by how elegantly it guides us through a workflow, but by how much of that workflow we no longer need to perform ourselves.

The design problem changes from "How do we help users complete this workflow?" to "How do we let software carry out the workflow while keeping users informed and in control?"

The CMS isn't going away. Its role might be changing.

Statamic still matters. Its files and blueprints give content a structure: titles, bodies, images, tags, relationships, and metadata. An agent needs that structure to make changes safely and correctly.

APIs, permissions, and validation become more important, not less, when software acts on our behalf. The CMS remains the system of record, even if the conversation becomes the workspace.

One example is the Ideas feature I built for my articles. I decided what it should do and defined the fields in Statamic: a thought worth exploring, its author, and a link to the source. Now I can ask ChatGPT to research relevant outside perspectives and populate that structure directly in the CMS. What used to mean researching, copying links, and filling in entries myself becomes an editorial conversation. And yes, I used exactly that workflow to find the ideas accompanying this article. Pretty meta. 😄

That's the distinction I don't want to lose. Designing the bones of the site is still a human job — deciding what belongs, how information relates, and what makes the experience useful takes thought and care. Once that foundation is in place, AI is a fantastic editorial assistant for researching, organizing, and populating richer content. It doesn't replace the design decisions; it helps me make much better use of them.

Humans still own the outcome

I don't want an agent making every editorial decision for me. The opinions, experiences, and judgment still need to come from me. I want help shaping an argument, finding gaps, and reducing repetitive work, not a system that confidently publishes whatever it generates.

That means access should match the task, changes should be reviewable, and publishing should be a deliberate decision. A technically successful update can still contain a weak argument or an awkward sentence.

The same principle applies beyond content management. If agents become a common way to interact with software, trust, reversibility, and visibility into what actually changed will be part of the user experience, not just implementation details.

A small experiment with bigger implications

In The Full-Context Technologist, I wrote about giving AI the context it needs to help across tools and disciplines. Connecting ChatGPT to my blog is one small, practical example. What started as an easier way to write has changed how I think about using software.

And yes, this article was created through that very workflow. A little meta, but I couldn't resist. 😄

The other side of the conversation

There’s an irony here that I can’t ignore. I’m managing my blog through a conversation, and readers can now consume it through one, too.

You don’t have to visit my site to encounter my ideas. You can ask an agent a question and get an answer that draws on something I’ve written, with a citation pointing back here. My words can reach you without you ever seeing the experience I built around them. As an author, that’s both exciting and a little unsettling.

That also changes how I think about publishing. Alongside making content accessible to browsers and crawlers, there’s growing value in exposing it directly to agents through MCP. A server that lets connected agents search my published articles and retrieve their text, author, and original URLs gives them another route to the source.

And the version we expose to agents doesn’t need to be pretty. It needs to be clear, well structured, and easy to retrieve, with enough context to preserve the meaning and credit the source. The typography, animations, and carefully chosen layouts still matter to the people visiting my site. An agent has different needs.

The same content can serve both audiences without sharing the same presentation. That’s another shift in UX: designing how information is accessed and understood, even when there’s no page for someone to look at.

Publishing now means thinking about both the person visiting your site and the agent reading on their behalf. And if you’re reading this through an agent, you’re already part of that shift.