Umbraco's CodeGarden 2026: AI, automation, and what it means for your platform

This is my third year at CodeGarden, and the excitement of heading to Umbraco's main conference hasn't faded. This year the event made its return to Copenhagen, the city where it all began back in 2005. The venue, Øksnehallen, was perfectly nested in the city centre, and with around 700 attendees across two packed days, it had the energy of a community that's genuinely grown into itself.

For me personally, this year carried a little extra weight. Ahead of the conference I was awarded Umbraco MVP status for the first time in recognition for contributions to the Umbraco Leeds community (UmbLeeds) and development of my package. It's the kind of thing that reminds you why giving back to an open-source community matters; the people around you notice, and they push you to keep going.

As always, I've come back with a notebook full of ideas and a head full of things I want to explore. Here's what stood out.

Keynote Highlights

The opening keynote is always the moment where Umbraco HQ lays their cards on the table for the year ahead, and this one didn't disappoint. The overarching message was clear: Umbraco is doubling down on being a platform that both developers and the businesses using it can rely on.

What's Next for the CMS

With Umbraco 17 already well established as the current LTS release, the keynote's focus was firmly on what comes next. Umbraco 18 lands on 25 June and is modest in scope, with one headline feature arriving in its first phase and a handful of focused improvements under the hood. The most developer-facing change is the swap from Swashbuckle to Microsoft's own OpenAPI library. Most people won't notice, but if you maintain packages or Backoffice extensions, it's worth testing against now.

Looking further ahead, Umbraco 19 brings Entity Framework support. Moving away from NPoco, this opens the door to database providers beyond SQL Server and is hugely relevant for enterprise projects with specific infrastructure requirements.

Reusable Content and Smarter Search

Two features stood out as genuinely significant for how we build and manage content in Umbraco going forward.

Elements is a new content type built around the idea of reusable blocks and components that can be shared across grids and lists without content being duplicated across the tree or requiring workaround solutions. For editors it means a new dedicated library section without the copy-paste headache. For developers it means content models that are cleaner and considerably easier to maintain over time.

Search has been rethought from the ground up. Rather than being tied to a single provider, Umbraco is introducing an abstraction layer so teams can choose search providers based on what a project needs. Faceted search, field-level querying, and more flexible indexing are all part of what's coming, the kind of change that opens proper search experiences without requiring bespoke solutions. This will first be released as a package for Umbraco 17 before moving into the core CMS with version 18.

Real-time editor collaboration is also on the horizon, which will be a welcome change for content teams working across the same site simultaneously but still in the early stages of development.

AI on Your Terms

The AI section of the keynote was the most substantial, and refreshingly, the emphasis wasn't on AI as a buzzword. The consistent thread was control, giving users the choice of which models to use, what they can touch, and having safeguards in place before anything reaches your content.

On the governance side, new guardrails mean prompts and AI output can be inspected before or after generation, helping keep content on-brand and ensuring sensitive data doesn't get passed to an external model unnecessarily. Testing capabilities lets you benchmark AI output, compare how different models perform on your specific content, and validate against mock data before going anywhere near a live environment. For clients who've been cautious about AI, this kind of tooling makes the conversation a lot easier.

Furthermore, the Copilot assistant for editors is now released. Editors can dictate content, drag files straight into the chat window, and use it to find, create, and update content including within complex block structures. Crucially, there's always a manual approval step and nothing publishes without a human signing off.

Bringing Umbraco into Your AI Workflow

MCP (Model Context Protocol) was a big talking point last year, and this year Umbraco has taken it significantly further. There are now two distinct offerings: the existing Developer MCP for working with the Management API, and a new Editor MCP that can be hosted directly on Umbraco Cloud. This means editors can connect tools like Claude or other AI clients and manage content through natural language without anything needing to be installed locally. Both will be available on all Cloud plans at no extra cost, with beta arriving in July.

A new MCP CLI tool rounds this out for developers, and pre-built Agent Skills cover common tasks like working with content models, so there's a practical starting point rather than having to build everything from scratch.

Umbraco Compose: Composable for Real

Compose has been quietly building momentum since its launch earlier this year, and the keynote provided a chance to reflect on where it's heading. For those unfamiliar, Compose is Umbraco's data orchestration platform, sitting in the middle layer between your various data sources and your frontend experiences. It ingests content from multiple systems through a managed API, standardises it, and delivers it via GraphQL, meaning frontend teams get clean, structured data without the usual custom integration headaches.

The updates shown were focused on giving teams more control and visibility: a new UI for managing API applications and scopes, ingested content diagnosis so you can see exactly what's been pulled in and how it maps to your schemas, outbound webhook functions for manipulating what gets sent to external services like search engines, and more granular ingestion management. For enterprise projects with complex data environments, Compose is increasingly the answer to the question of how you make a composable architecture work in practice rather than just in theory.

Automation Without Leaving Umbraco

The announcement that genuinely caught the room's attention was Umbraco Automate, a brand new open-source add-on that brings workflow automation natively into the Backoffice. Automate provides a visual drag-and-drop interface without leaving Umbraco and without sending your data through a third-party automation platform.

It works across the CMS, Forms, Commerce, Workflow, and Engage out of the box, supports webhooks for connecting external services, and can incorporate AI steps into flows. Importantly, human approval steps can be built in so automated processes don't bypass editorial oversight. It's in beta for Umbraco 17 now with a full release on 9 July, and being open source means the community will inevitably extend it quickly.

The Rest of the Ecosystem

The keynote also ran through updates across the other key products. Engage gains MCP and Copilot integration. Forms gets a new analytics section and MCP support. Commerce introduces a customer portal with planned MCP support. Deploy has been refreshed with quality-of-life improvements.

Finally, Cloud picked up arguably the most practically useful set of changes: better insights, improved error visibility, improved carbon reporting, hosting recommendations, and most significantly load balancing with Redis cache support. Furthermore, scheduled updates to give stronger control on when automatic upgrades should occur, which will help avoid peak traffic times.

Sessions and Talk Highlights

From Prompts to Production

One of the most entertaining and genuinely useful talks of the conference was Matthew Wise talk “Vibe Coder to AI Engineer”. Matt walked through his journey of using Claude Code to build a fully CMS-managed Umbraco experience from designs through to content modelling and the finished product. Refreshingly, it wasn't a highlight reel. The demos included moments where Claude went off in unexpected directions, prompts fell flat, and the classic "it works on my machine" made an appearance.

The distinction Matt drew that stuck with me was the difference between a vibe coder and an AI engineer. A vibe coder throws prompts at a problem and hopes for the best; it works for them in the moment, but it isn't shareable or maintainable. An AI engineer treats AI as part of a real workflow, breaking work into well-structured files and folders, using context engineering techniques like planning documents and memory, and knowing when to step in and give the AI a firm steer. The tools are genuinely powerful, but the skill is in how you set them up to succeed.

Cutting Through the AI Noise

Richard Campbell, best known for hosting the .NET Rocks podcast, brought a broader perspective to the AI conversation with his talk “After the AI Hype: What’s Real, What’s Next” that was a welcome contrast to the usual AI hype. Rather than focusing on what AI can do right now, he placed the current moment in a longer historical context. AI has been through hype cycles before, and many of the lofty promises made during those periods didn't survive contact with reality.

That said, Richard wasn't dismissive. His point was less "AI is overblown" and more "learn to tell the difference between the noise and what's actually delivering value". For anyone making decisions about where to invest in AI tooling, it was a grounding talk that's worth keeping in mind as the market continues to mature.

Building for When Things Go Wrong

Callumn Whyte’s talk “Building Resilient Umbraco Integrations” was easy to overlook but important for anyone running a site that connects to external services. Callum’s core argument was simple: failure in software is inevitable, so the question isn't whether something will go wrong, it's whether your system is built to handle it gracefully when it does.

For developers, the practical takeaways centred on patterns like retry logic with backoff strategies and circuit breakers using Polly.NET as a real-world example. Callum also provided a clear breakdown of Umbraco's caching layers: request cache for the lifetime of a single request, runtime cache for the life of the application, and isolated cache for scoping to a specific service. Understanding which layer to reach for in each situation makes a meaningful difference to both performance and resilience.

The broader point though is if your site integrates with third-party APIs, whether that's a CRM, payment provider, or anything else, how those integrations are built determines whether a downstream outage takes your site down with it or whether your users simply don't notice.

Is Your Site Ready for AI Search?

One of the most thought-provoking talks of the conference, and one with direct implications for every site we build. Kyle Brigham talk “The Future Is AI Search: Designing Umbraco for AEO and AI Discovery” argued that the era of optimising purely for blue links on a Google results page is fading. Search is now an ecosystem, not a destination, and if an AI answers a user's question directly without sending them to your site, the only way you benefit is if you're the source being cited.

The reframe that landed hardest was this: when someone says, "I want to show up in search", what they usually mean is they want structured, well-organised content that AI systems and search engines can understand and trust. The Backoffice you build sets the ceiling on content quality and if the structure isn't there, editors can't produce content that performs.

Practically, Kyle walked through four shifts needed to adapt: entity thinking, content depth, making content citable, and accommodating how LLMs consume information. Every page should be able to answer the question of what it is the answer to. He also spoke of Answer Engine Optimisation (AEO) as a checklist-driven approach, covering things like LLMs.txt, FAQ structured data, schema generated per document type, page titles framed as questions, internal linking that reflects topic clusters rather than just navigation, and meaningful alt text on every image.

For us, this reinforces something we already believe: a well-architected Umbraco Backoffice and a strong content strategy will always outperform any SEO plugin.

Looking Forward: Key Implications

CodeGarden 2026 left me with a clear sense of where Umbraco is heading and what that means for the work we do at MSQ DX. A few themes cut across almost every session and announcement:

AI That's Actually Governable. The shift from "AI as a feature" to "AI as a governed, controllable layer" was everywhere this year. Between guardrails, testing capabilities, and the human-in-the-loop approach baked into both Copilot and Automate, there's now a credible answer to the question clients have been asking: how do we use AI without losing control of our content and data? That conversation just got a lot easier.

Structure Is Everything. Whether it was Kyle Brigham making the case for AEO-ready content models, Matthew Wise demonstrating how well-structured files make AI-assisted development work in production, or Elements giving content a proper home outside the document tree, the message was consistent. The decisions made at the architecture and content modelling stage determines everything that follows, both for human editors and AI agents working with your content.

Build for Failure, Not Just for Success. Callum Whyte's talk was a timely reminder that resilience isn't a nice-to-have. As sites become more connected through integrations, APIs, and AI services, the question of what happens when something goes wrong matters more than ever. Building with patterns like circuit breakers, retry logic, and appropriate caching isn't over-engineering; it's just good practice.

Search Is No Longer Just Search. The combination of Umbraco's new search abstraction, AI-powered semantic search, and the AEO conversation signal a fundamental shift in how we think about content discoverability. Optimising for AI-driven answer engines is quickly becoming as important as optimising for traditional search rankings.

These aren't abstract ideas for future projects. They're things we're factoring into how we approach Umbraco builds at MSQ DX, from how we structure content models and integrations to how we advise clients on AI readiness.