AI agents in Umbraco: how to use them and what you get out of it

An AI that fills in a landing page by itself in our CMS. A year ago, that sounded like a distant promise. Today it works. Umbraco now has its own AI assistant: an agent that works alongside you in the backoffice, creating and editing content for you. That changes what a marketing team can handle on its own, without pulling in a developer every time.

We'll explain what an AI agent in Umbraco actually is, how it differs from other AI features, what it delivers for your team, and why the way you set it up decides whether it truly works for you.

What is an Umbraco AI agent?

An AI agent isn't just a chatbot that spits out text. It's an AI that carries out multi-step tasks in the backoffice on its own. You give it an instruction in plain language, and it turns that into concrete actions: building a page, filling in blocks, rewriting a text, or linking media.

In Umbraco, the agent lives in a chat window next to your content, often called the Copilot. You type, for example, “create a new services page about maintenance,” and the agent thinks it through, picks the right steps, and carries them out. The difference from a standalone text generator is in that last step: a chatbot hands you text to paste in yourself, while an agent does the work in Umbraco itself. Think of the difference between a colleague who gives you advice and a colleague who also does the job.

That's possible because the agent runs on top of Umbraco's AI layer. This layer controls which AI model is used, what information the agent can see, and the limits it works within. So the AI doesn't work blind. It works within the boundaries of your own website and brand.

Agent, prompt or integration: what's what?

A few terms float around when it comes to AI in Umbraco, and they're easy to mix up. Getting them straight helps you make the right choice.

The simplest form is the one-click prompt. That's a button next to a field that lets you generate, for example, an SEO title or alt text with a single click. Handy for small, isolated tasks.

One step further is the AI agent, or the Copilot in the backoffice. It carries out bigger, multi-step instructions and can build an entire page. This is what this article is about.

Then there's MCP, the Model Context Protocol. That's something else. It's a connection that lets an AI tool outside Umbraco work with your website, such as your own AI assistant. In short: the agent works inside the backoffice, while MCP connects the outside world to your site. For the day-to-day content work of a marketing team, the agent in the backoffice is the most direct route.

What does it deliver for your team?

For a marketing team, the value lies mainly in time and speed. The tedious, repetitive work moves to the background, so you can focus on strategy and creativity.

A few concrete examples. Ask the agent to set up a new campaign page, complete with draft copy and the right product images. Have it rewrite existing pages so they match your tone of voice. Accessibility gets easier too: think of automatically filling in alt text or spotting missing headings. In our blog on AI in Umbraco CMS, we already showed how many smart applications there are.

The result isn't that AI takes over your work. The result is that your smaller team can handle more, switch gears faster, and spend less time on manual clicking. That's exactly what many marketing departments run into when the content schedule gets busy.

Why the right settings make all the difference

This is the heart of the story. An AI agent is only as good as the boundaries and context you give it. Set it up carelessly, and it gets expensive, slow, or delivers half-finished pages. Set it up well, and it works like a reliable colleague.

The profile: model and behavior

It starts with the profile. This is where you define which AI model the agent uses and how it behaves: whether it can be creative or should stay strictly within the lines. For one type of task you want a powerful, extensive model, for another a faster and cheaper one. That choice determines both the quality and the cost of every request.

The context: the right information at the right moment

Just as important is the context. An agent only performs well if it has the right information in front of it at the right moment:

• your tone of voice
• how a page is structured
• which blocks exist
• what is and isn't allowed

You define this in contexts and in the instructions you give the agent. Give it too little, and the agent starts guessing and the page won't be right. Give it the right context at the right moment, and it fills in a page that looks right straight away. This is often the difference between an agent that impresses and one that disappoints.

Actions, tools, and guardrails

You also decide which actions and tools the agent may use. You give it targeted permissions to, for example, rewrite content, create a page, or save changes, and nothing more than that. With guardrails, you also set hard limits so the agent never colors outside the lines. That's not just a matter of safety, but of cost too: fewer active tools means fewer tokens per request, and so lower costs and faster answers.

Human in the loop

Finally, keep a human in the loop. Let the agent make suggestions and deliver drafts, but leave publishing to an editor. That way you combine the speed of AI with the judgment of your team.

We help organizations with exactly this setup, from profiles and context to permissions and smart working agreements, so AI in your Umbraco website does what you want and nothing more.