Could you manage your websites content from just one input field?

Dan Smith's avatar

Dan Smith

Published: February, 2026

One Agent, Many CMS - Managing multiple content management systems

TL;DR

Instead of choosing a CMS based on familiarity, focus on storage, structure & cost and build an agent-driven interface that works with any CMS backend. We built a prototype where a single text field manages website content through natural language requests. After testing, we see strong evidence this approach excels at post-live updates, converting human language into content changes faster than traditional CMS dashboards.

Try the prototype here.

When it comes to choosing a content management system for a website, should we use Drupal, WordPress, or perhaps a SaaS-based one like Contentful or Sanity?

When working with creative directors, marketing/IT managers, commonly an influencing factor is the user interface.

How can we make this choice simpler? We start by establishing the type of product this CMS is powering—from a simple marketing landing page to a large corporate enterprise. I don't believe we should be restricted by the user interface. Instead, we should focus on storage, structure & cost.

With this influencing factor out of mind, does this help simplify your choice? Consider what you're actually storing, how it needs to be organised, and what it will cost. If the content is mission-critical, powering multiple tech products, we can focus on a SaaS content manager—some even come with their own AI content managers.

Maybe we're working with something more concentrated, perhaps a simple landing page, blog, or marketing website. Maybe your data is rightfully stored across multiple systems?

This is the crux of the question: maybe there's some space here for a simple interface that sits in the middle? One text field backed by a content management agent (or team of agents) that evaluates what you'd like to do and executes it if clear.

The minimal viable product

To test this concept, we'll build a simple prototype that demonstrates how a content management agent can work with a basic content structure.

For the MVP, we'll focus on the core functionality:

  • When the visitor visits the editor page, a duplicate of the site is generated
  • A single text field where users can input content management requests overlayed over the duplicate site
  • A content management agent that interprets the request and determines the appropriate action
  • A simple content structure that can be easily updated and parsed

With this in mind, we can now build the prototype.

  • Build a simple static sitelinelinelineWe'll build a simple React Vite application that will show a simple landing page. with a single page
  • A well-defined content structurelinelinelineWe structure the content into .json files and parse those into the components. The filename of each .json file represents the last part of the URL, so each .json file we create automatically maps to a page. For now, we'll keep it simple with just one page to simplify the logic for the agent. that maps directly to page components
  • We'll keep the page count to one page for now to maintain simplicity
  • Basic agent capabilities that can understand simple content update requests

The first step is to construct a rapid prototype to obtain an idea of the scale of work. We'll begin with the most basic system. Commonly when building a web product, if there's no specific desire for a CMS, my default approach is to make sure the content structure is well-defined and can be easily migrated or adapted to any CMS platform later.

The prototype

Here's a walkthrough of the prototype we built: Try the prototype here.

The user begins their journey with being presented with the “Setting up your website instance” screen. Behind the scenes, we're making a clone of the currently live website so it's safe to edit.
Presenting the user with some basic privacy notification. The user's conversation will be stored, some basic tracking data & the files they upload will be stored.
We initiate the user experience by adding an opaque layer above the website in question with a CTA drawing the user's eye to the edit button.
In this example, we're going to request that the management agent updates the hero image. We can just literally ask the agent “Please update the hero image”
The agent will identify the request requires a new image to be provided and will provide the user with an option to upload a new image
Drag and drop the new image into the overlay and the agent will update the hero image on the website
Wait for the upload to complete
Let's change the title underneath the hero image. We prompt the agent to “Please update the title underneath the hero image to “The CMA Prototype”
We see that the title has been updated.
Let's prompt the CMA to remove the entire get started section of the website.
And it's gone!
  • The user begins their journey with being presented with the Setting up your website instance screen. Behind the scenes, we're making a clone of the currently live website so it's safe to edit.
  • Presenting the user with some basic privacy notification. The user's conversation will be stored, some basic tracking data & the files they upload will be stored.
  • We initiate the user experience by adding an opaque layer above the website in question with a CTA drawing the user's eye to the edit button.
  • In this example, we're going to request that the management agent updates the hero image. We can just literally ask the agent Please update the hero image
  • The agent will identify the request requires a new image to be provided and will provide the user with an option to upload a new image
  • Drag and drop the new image into the overlay and the agent will update the hero image on the website
  • Wait for the upload to complete
  • Let's change the title underneath the hero image. We prompt the agent to Please update the title underneath the hero image to The CMA Prototype
  • We can ask the agent to remove entire sections of the website.
  • Let's prompt the CMA to remove the entire get started section of the website.
  • And it's gone!

Reflections and next steps

When the CMA does its job correctly, it really does seem like a viable complement to any CMS for post-live tweaks and updates. It can also be used as a lightweight CMS in its own right. It feels quick, and the AI agent seems mostly competent—making it easier than logging into the dashboard for simple content changes.

However, the next phase comes with many more challenges. Currently, the content management agent just reads a plain text file and updates it with the relevant content. There are a host of future improvements and challenges to overcome to potentially be a viable product for the end user.

  • How will the changes be deployed live safely?
  • How do we safeguard against wider unintended edits from the CMA?
  • What should the initial CMS offering be?

This is somewhat of a larger undertaking, but there could be space for such a product.

Need a hand?

Ad-hoc planning, prototyping, building, testing, maintenance and team management in all aspects of web development. No upfront commitment and an initial free consultation, check out our process or let's chat and see how we can help.

Company logo for Websultancy abbreviated as WS wrapped in HTML opening and closing tags.