Umbraco Codegarden 2026 event highlights and AI roadmap

Codegarden 2026: Umbraco's AI roadmap, and what it really means for our clients

Last week the NXT team spent four days in Copenhagen at Codegarden 2026, Umbraco's flagship conference and the largest annual gathering of the global Umbraco community. We were on the ground from Tuesday 9th to Friday 12th June and this year's event was particularly special for us.

Ahead of the conference, NXT was shortlisted for the Best Innovation with AI award at the Umbraco Awards 2026 in recognition of our work helping Exclusive Networks bring AI into their digital platforms in a practical, governed and measurable way. Being shortlisted alongside some of the most innovative agencies and partners in the Umbraco ecosystem was an achievement in itself and a proud moment for the whole team. But we weren't in Copenhagen just to celebrate. Like any good Umbraco partner, we were there to get the roadmap straight from the people building it, pressure-test it against the real problems our clients are trying to solve, and come home with a clear view of where the platform is heading next.

The headline from the keynotes weres that Umbraco has worked out how a serious enterprise should do AI inside a content platform: in the open, under governance, with a human in control. That's a very different proposition from other CMS solutions.  If you want the official version, Umbraco's CTO Filip Bech-Larsen published a thorough keynote highlights post on the day. Here's NXT's highlights. 

That was the line Filip Bech-Larsen returned to throughout the product keynote, and it's a good summary of where Umbraco has positioned itself. The last few years have been about credibility at the enterprise level — ISO 27001 certification, a 100% SLA commitment, strong independent reviews — and this year's announcements were all consistent with that direction. Maturity, not novelty, was the through-line.

For an agency like ours, that matters. We build for complex, often global organisations in finance, technology, engineering and cybersecurity. Those clients don't want to be anyone's beta test. When a platform vendor leads with trust, governance and operational stability and then layers AI on top of it, that's a conversation we can take into a boardroom. The roadmap shared on stage covered the next twelve months across the whole suite, and it broke cleanly into two stories: the platform getting more capable, and AI being embedded the Umbraco way.

The platform story: Umbraco 17 to 18, and what's coming next

Umbraco 18 lands on 25th June 2026. Reassuringly, it's a small, low-friction major release, most projects won't need significant re-work to upgrade. But it introduces a couple of things worth planning around:

• Elements — a new Library section that lets editors create a piece of content once and reuse it across as many pages as they like. Calls-to-action, contact details, opening hours, legal disclaimers, author bios: create once, update once, and it changes everywhere. It's a genuine single source of truth for repeated content, and full Block Editor integration (reusing blocks from the Library) is following later in the year. We've been doing our own version of this for a few years utilising Web Settings but good to see this coming. 

• Umbraco Search abstraction (final release) — a provider-based search architecture that gives developers one consistent way to express advanced search (full-text, filtering, faceting, sorting, language and segment support) while keeping the freedom to plug in the search engine of their choice. Examine stays the default, but if a client needs Elasticsearch or Typesense, the stable API is now there to build against.

Looking further out, Umbraco confirmed two foundational shifts: a move from the NPoco micro-ORM to Entity Framework Core (targeted for Umbraco 19), which brings Umbraco into line with standard modern .NET data access and makes testing faster and more reliable; and the longer-term plan to bring Umbraco Search into the CMS core. None of this is glamorous, but it's exactly the kind of under-the-hood investment that keeps a platform healthy for the decade-long lifespans our enterprise builds tend to have.

AI in Umbraco

Here's where it got interesting. Umbraco split its AI narrative into two halves, and the distinction is more than semantics.

The first half is AI in Umbraco — bringing AI capability into the back office, on Umbraco's terms. Three things stood out:

Governance built for platform owners. New guardrails inspect both the prompts going into a model and the output coming out of it — before or after generation — so content stays on-brand and sensitive data stays protected. Crucially, you choose the model: new connections include Hugging Face, Fireworks, Mistral and DeepSeek alongside the established providers, so the LLM can be matched to a project's budget, performance and compliance needs rather than dictated to you.

A test-and-run capability. You can now measure agent and prompt output against defined standards — Umbraco calls them "graders" — to catch model regressions, compare models, and run against mock data before AI ever touches live content. In plain terms: you can quality-check the AI in your back office the way you'd quality-check any other part of a release. For regulated clients, that's the difference between "we use AI" and "we can evidence how our AI behaves."

A Copilot for editors, now in final release, with speech-to-text dictation, drag-and-drop file uploads, semantic search, support for complex block structures, and — the part that matters most — a manual approval step so an editor always signs off before anything publishes.

That phrase "human in control" came up again and again. Umbraco even shipped a security patch for the Umbraco.AI package in the run-up to Codegarden and talked about it openly on stage. That's the right instinct, and it's the instinct our clients want from a platform.

Umbraco in AI — MCP and Agent Skills

The second half — Umbraco in AI — is the one we're most excited about, because it's where Umbraco becomes a first-class citizen inside the AI tools teams already use.

The headline is the Umbraco Editor MCP: a remote Model Context Protocol server that Umbraco hosts for you on Umbraco Cloud. In practice, an editor opens their preferred LLM client and can draft, find, update and publish content directly in the Umbraco back office just by describing what they want. There's nothing to install or maintain — Umbraco hosts it, each Cloud project gets its own authenticated (OAuth) endpoint, and it's already covered by the existing Umbraco Cloud data processing agreement. Both the CMS Editor MCP and the CMS Developer MCP are landing as hosted remote MCPs on Umbraco Cloud this summer (beta in July, general availability later in the summer), at no additional cost on any Cloud plan.

Alongside the MCP work, Umbraco introduced Agent Skills — supported, maintainable building blocks that let AI agents do real development and configuration work without bespoke, throwaway integrations. The four skill areas shared on stage were:

1. Back-office extension skills — building dashboards and back-office UI, including help migrating extensions from Umbraco 13 through to 17. Implementation skills — project setup and scaffolding.

2. Content modelling skills — defining and managing content structures.

3. Cloud skills — setting up and configuring Umbraco Cloud environments.

The reason this is strong, in Umbraco's own framing, is that it's inspectable and extendable. The CMS now exposes its content structure to agents so they can read exactly how a page is built before changing anything — and make precise, granular edits (even deep inside a Block List) instead of clumsily rewriting a whole page. That's a structural advantage, and it's open to any agent, not just Umbraco's own Copilot. If you'd rather bring your own LLM or tooling, you can. For our developers, that openness is the whole game.

Umbraco Automate: the new product nobody quite predicted

Umbraco also launched a brand-new, open-source automation add-on: Umbraco Automate. It's a drag-and-drop automation engine native to the back office — the "if this, then that" flows you'd normally wire up in an external tool, now running where your content actually lives, built on the same triggers-and-actions model already familiar from Forms and Engage.

What makes it interesting for our clients isn't the convenience — it's the data posture. Unlike Zapier or Make, Automate runs entirely inside your own Umbraco infrastructure, whether that's Umbraco Cloud, a dedicated Azure SQL database or self-hosted. Your data only leaves your environment when you choose to send it somewhere. You can hook into external services via webhooks, call Umbraco.AI agents as steps within a flow, and keep human-in-the-loop approval before anything external or AI-generated goes live. It's open source, endlessly extensible, with a beta available now on Umbraco 17 and a final release on 9th July. You can read more on the Automate product page.

What this means in real life

This is the part I came home wanting to write down, because it's easy to nod along to a roadmap and miss the practical shift underneath it.

We've been working with Umbraco's custom MCP for a while now. What these new extensions unlock is a properly joined-up, governed way of orchestrating AI across a client's whole stack — without giving up the security and control that the CMS provides out of the box. Here's a concrete, end-to-end example of the kind of workflow we can now design:

1. Stand up a customised MCP for the client's Umbraco instance and set the context the AI is allowed to operate within — brand voice, tone of voice, and the data-governance rules that decide what the model can and cannot see or touch. This is the guardrail layer: we prohibit and manage exactly how any LLM interacts with the data we're prepared to share.

2. Bring in other MCP integrations — for example SEMrush for organic and GEO performance data, and HubSpot for CRM and conversion data — into the same governed workflow.

3. Write a prompt that becomes an action. Something like: "Identify our top-performing landing pages that attract returning visitors, replicate the structure of the best one, and generate fresh copy for a new page based on the brand and product guidance in this shared document."

4. The agent interrogates the data, makes suggestions, drafts the copy from the client's own insight and a controlled source document held in their online drive, and assembles the new page as content blocks directly in the back office — ready for an editor to review, refine and approve.

That's the revolution: integration and joined-up management of the CMS, with the AI doing the heavy lifting across analytics, CRM and content, all while the security and governance stay anchored in the platform. And this is just the step-by-step version. The same building blocks let us develop a standing internal agent for a client — one you can be explicit with, train on facts and responsibilities, and task with managing processes as well as attempting to solve them. You can wrap tools and LLMs together while still being granular about write permissions. That's a meaningful step beyond "AI as a clever autocomplete."

It's why I think Umbraco is uniquely positioned to become the number one CMS of choice for AI. The combination of an inspectable, extendable content model, hosted MCP infrastructure, model choice, guardrails and human-in-the-loop approval is not something the proprietary DXPs can easily match. Open source plus governance is a genuinely differentiated place to stand.

Umbraco Awards

After the first day of keynote sessions, the community gathered for the annual Umbraco Awards. Having already been shortlisted for Best Innovation with AI, we were looking forward to celebrating the achievements of teams across the ecosystem. We were therefore delighted — and genuinely surprised — to hear NXT announced as the winner. You can read the full story in our dedicated post: NXT Wins Umbraco Awards 2026. The recognition was a fantastic endorsement of the work our team has been doing around practical, governed AI solutions, and it rounded off an exceptional first day in Copenhagen before the focus shifted firmly back to the roadmap, product announcements and conversations that will shape client platforms over the coming year.

The wider picture: the dates that matter

It's easy to leave a conference with a head full of vision and no sense of timing — so here's the orientation that counts, because several of these announcements become real releases within weeks, not years:

• Umbraco 18 — final release 25 June 2026 (Elements Phase One, plus typed OpenAPI schemas for the Delivery API)

• Umbraco Cloud Load Balancing — 1 July 2026

• Umbraco Automate — final release 9 July 2026 (open-source beta on Umbraco 17 available now)

• Editor and Developer MCP — beta in July 2026, general availability later in the summer

And the momentum hasn't slowed since we flew home. The Umbraco 18 release candidate is already in community testing ahead of the 25 June launch, and early reaction has been warm — developers are singling out Elements and the new typed OpenAPI schemas as the changes that make content modelling cleaner and headless builds smoother.

Author: Peter Hinchliffe is Strategy Director at NXT Digital Solutions, a Guildford-based Umbraco development agency and Microsoft Partner. NXT builds enterprise Umbraco platforms and AI agent solutions  for complex, global organisations.

Supporting links

• Umbraco — Codegarden 2026 keynote highlights 

• Umbraco 18 release candidate (final release 25 June)

• Umbraco Automate product page

• Umbraco AI governance

• Umbraco Developer MCP

• The Agent-Ready CMS — what are Agent Skills?

• Umbraco product roadmap

• Umbraco release schedule

• NXT — Best Innovation with AI win at Umbraco Awards 2026