A representation of agentic CMSs, generated by Gemini.

Every CMS is agentic now. So what does ‘agentic CMS’ even mean?

Every major DXP and CMS now ships an ‘agentic’ feature, and the vendors mean wildly different things by it. A view from the agency side on what an agentic CMS actually is, who’s making which bet, and the one question that decides which platform ages well.

Adam Griffith

21 July 2026

15 minute read

The short version

  • An agentic CMS can act on a goal, not just assist with a task. The dividing line isn’t whether it has AI features; it’s whether it can take actions you didn’t individually trigger, inside guardrails you set.
  • Six platforms we work with are making six different architectural bets. The real difference between them is where the agent runtime sits: inside the CMS, above the stack, in your own tools, or with a different vendor entirely.
  • Mid-2026 made the trend impossible to miss: Optimizely rebranded as ‘the AI platform for marketing’, and Salesforce moved to buy Contentful as a content layer for its agents.
  • AI is now part of the floor, not the tie-breaker. The new selection question is whether you can own the agent layer or at least leave it, and MCP support is the cleanest proxy for that.


On this page: 

  1. What is an agentic CMS?
  2. We’ve seen this hype before
  3. Six bets, one market
  4. What does ‘table stakes’ look like now?
  5. The AI-visibility land grab
  6. The fight that actually matters
  7. Five questions worth asking
  8. What we tell clients now
  9. Frequently asked questions

When I wrote about AI inside the CMS early last year, I said the features turning up in every platform would be table stakes within a year. They are.

Every major DXP and CMS we work with at Luminary, from Optimizely and Sitecore through to Kentico, Kontent.ai, Contentful and Umbraco, has shipped some flavour of agentic feature in the last 12 months.

The pitch decks have moved on, and so has the buzzword. We’re done with ‘AI-ready’. The phrase now is ‘agentic CMS’.

Except half the vendors saying it mean wildly different things by it. And the sharpest critic of weak agentic claims is, awkwardly, a vendor selling agentic features of its own.

So here’s the follow-up I owe you. Not a vendor scorecard. A view from the agency side on what an agentic CMS actually is, what’s really shipping, what’s marketing fog, and the one question worth asking when you choose a platform in the back half of 2026.

What is an agentic CMS?

Start with the word, because the word is doing a lot of work.

Definition. 

An agentic CMS is a content management system that can act, not just assist. Where an AI-assisted CMS suggests a headline or drafts a summary for you to accept, an agentic CMS can take a goal (‘localise this campaign for five markets’), break it into steps, use tools and other systems to do the work, and carry it out with limited human input, inside the guardrails and approvals you set.

That last clause matters as much as the first. Agency without guardrails isn’t a feature. It’s a liability.

It helps to see it as a spectrum, not a switch:

  • Assisted. The editor asks; the AI answers. Autocomplete, ‘improve this paragraph’, alt text on request. Useful. Not agentic.
  • Automated. Fixed rules fire on triggers. Schedule a post, route for approval, tag on upload. We’ve had this for years. Also not agentic.
  • Agentic. The system works towards an outcome you’ve described, decides the steps itself, calls the tools it needs, acts, and then tells you what it did. That’s the line.

The sharpest version of that distinction came, of all places, from a vendor. Optimizely argued that calling a CMS agentic because it drafts faster is ‘like calling a car autonomous because it has cruise control’. When the best line about your category is coming from inside the category, the category is fighting the wrong fight.

Agentic CMS vs headless CMS vs a traditional CMS

These get muddled, so here’s the honest one-liner for each. A traditional CMS stores your content and waits for you. A headless CMS does the same, then serves that content to any front end through an API. An agentic CMS adds a third thing on top: a layer of software that can read your content, reason about a goal, and do the work.

Which is why the platforms best placed for this tend to be the ones with the most structured, well-modelled content underneath. Agents can’t act intelligently on a pile of page-shaped blobs. The better your content model, the more an agent can actually do with it. That detail decides more platform selections than the demo ever shows.

We’ve seen this hype before

Whenever the CMS category gets a new noun, the noun does the work of differentiation for a while, until the architectures underneath catch up and the noun stops meaning much. Headless. MACH. Composable. DXP. Each one was the unmissable label for two years, then quietly turned into a feature on everyone’s slide.

‘Agentic CMS’ is the 2026 version of the move.

And here’s the tell that we’re deep into the cycle: in June 2026, Optimizely dropped the DXP framing altogether and rebranded as ‘the AI platform for marketing’. When a category leader stops calling itself a DXP, the noun has stopped doing the work. The same research it published put numbers on the unease: 64 percent of B2B marketing leaders worry that mass AI adoption is creating a ‘sea of sameness’ (Optimizely-commissioned, so take the figure with the usual pinch of salt, but the direction rings true).

Six bets, one market

Peel the labels off and the six vendors we deal with most are making six genuinely different architectural bets. Worth a quick look at each, with the catch attached to each one.

Optimizely (Opal) – the orchestration bet

Opal sits above the Optimizely platform (CMS, CMP, Commerce, Experimentation, Personalization) as an agent runtime, not as a CMS feature. The June 2026 update added a Quality tab with custom evaluation criteria and a guardrail that learns an agent’s patterns, flags deviations and auto-terminates rogue runs, plus sharable canvases with granular permissions, Word and PowerPoint authoring, a GitHub MCP connector and EU hosting. The rebrand reframes all of it as the ‘Optimizely Agent Platform’, with a ‘marketing harness’ of accumulated brand context and history feeding every agent.

Watch-out: Opal’s value compounds when you run more than one Optimizely product. Run only the CMS and a lot of that orchestration story is waiting on the rest of the suite.

Sitecore (SitecoreAI, Agentic Studio, Stream) – the embedded workspace bet

Launched globally on 10 November 2025, with all XM Cloud customers upgraded automatically. It ships with 16 out-of-the-box agents, including an AEO and SEO Researcher that audits your visibility across ChatGPT, Copilot, Perplexity and Gemini alongside Google. Brand Kits ground the agents in your voice.

Watch-out: It’s tightly bound to XM Cloud plus Stream. The agents are Sitecore’s, the brand context lives in Sitecore, and the orchestration story for non-Sitecore systems is thinner than Opal’s.

Kontent.ai (Agentic CMS, Expert Agents) – the headless-agent bet

The vendor that put the term on the map, with the ‘world’s first agentic CMS’ line at its October 2025 launch. Expert Agents followed on 31 March 2026, covering governance, compliance, translation and large-scale updates, with externally audited AI governance (ISO/IEC 42001) as a genuine differentiator for regulated clients.

Watch-out: Its strength is autonomous content operations inside a headless CMS, which is also its narrowness. No commerce, no experimentation, no personalisation underneath.

Contentful (AI Actions, then Salesforce) – the assembly-layer bet

AI Actions has been generally available since 31 March 2025: embedded utility, not autonomous agents. The bigger news is the 1 June 2026 agreement for Salesforce to acquire Contentful, positioning it as the content layer for Agentforce.

Watch-out: The deal hasn’t closed (expected in the third quarter of Salesforce’s 2027 financial year), so the Agentforce integration is roadmap, not a shipped capability. Buy in mid-2026 and you’re implicitly buying Salesforce’s AI direction, not Contentful’s. More on that below.

Kentico (KentiCopilot, AIRA) – the dual-track bet

Two distinct products. KentiCopilot is developer-facing, with a Content Types MCP server and AI-assisted code-upgrade tooling. AIRA is the marketer-facing agent inside Xperience by Kentico, with an Agentic Marketing Suite that includes an SEO and GEO Specialist.

Watch-out: A smaller agent surface than Opal or SitecoreAI, with no orchestration story to match. The bet is ‘your existing dev tools, plus our MCP’. Pragmatic. Unflashy. Quite possibly the most quietly correct of the lot.

Umbraco – the MCP-as-platform bet

Of the six, Umbraco is the most architecturally honest. It isn’t shipping agents. It’s shipping the connective tissue agents need. The Q1 2026 product update outlines a base MCP SDK, Hosted MCP, a CMS Editor MCP and remote MCPs hosted on Umbraco Cloud.

Watch-out: No first-party agents. Anyone shopping for an out-of-the-box agentic experience will look elsewhere, and that’s the point.

Read those six side by side and a pattern shows up. The bets aren’t really about features. They’re about where the agent runs, and that resolves into four places.

Platform

The bet

Where the agent runs

MCP / openness

The catch

Optimizely (Opal)

Orchestration above the suite

Above the stack (suite vendor owns it)

Yes: CMS SaaS MCP server, remote MCPs (e.g. GitHub)

Compounds only if you run more than one Optimizely product

Sitecore (SitecoreAI)

Embedded agent workspace

Inside the CMS (XM Cloud)

Agents are Sitecore’s; thinner third-party story

Tightly bound to XM Cloud + Stream; context lives in Sitecore

Kontent.ai

Autonomous ops in a headless CMS

Inside the CMS UI

API-first; ISO 42001 governance

No commerce, experimentation or personalisation underneath

Contentful → Salesforce

Content layer for someone else’s agents

With a different vendor (Agentforce)

API-first; future tied to Salesforce

Integration is pre-close roadmap; you inherit Salesforce’s direction

Kentico

Dual-track: dev MCP + marketer agent

Your IDE/tools (KentiCopilot) and the CMS (AIRA)

Yes: Content Types MCP server

Smaller agent surface; no orchestration story to match

Umbraco

Ship the connective tissue, not the agents

Your IDE / your own LLM

Yes, by design (SDK, Hosted MCP, Editor MCP)

No first-party agents out of the box (deliberately)

What does ‘table stakes’ look like now?

A short audit. The 2026 floor for any serious enterprise CMS:

  • MCP support, in production or credibly on the roadmap.
  • An AEO or GEO agent built in (Sitecore, Optimizely and Kentico all ship one already).
  • A permission-inherited assistant that drafts, summarises and translates.
  • ‘No publish without human approval’ as a posted policy, not an aspiration.
  • Bulk content operations across many items at once.

If a vendor can’t match that list, they’re behind, whatever the deck says. The slide with the circle of agent icons does not count.

The same scramble is happening in AI visibility

That AEO or GEO agent sitting in the table-stakes list is the tip of its own land grab. While the vendors race to put agents inside the CMS, they’re racing just as hard to sell you tools that show how your brand turns up when an AI answers a question about you. In the space of a few weeks this month, Sitecore acquired its way in, Optimizely partnered and bolted on its own analytics, and Contentful bought a startup and relaunched it as a standalone product. Adobe had already made its move earlier in the year, and Kentico has been quietly threading a GEO agent through AIRA. Same conclusion, different bets, all at once.

It rhymes with the agentic story, catch included. These tools do earn their keep: they’ll catch a model confidently telling people something false about you, and show where a competitor is winning a topic you should own. What they can’t do is the half that actually decides whether an AI cites you, which lives out on the open web, in your reputation and authority, not on any dashboard. The agents will take the mechanical, on-site work off your hands. The half that sets you apart stays human.

I’ve pulled that argument apart in a companion piece: Every DXP is now selling AI visibility. They’ll do the easy half. The short version: buy the category if you like, but spend the real money where the tools can’t reach.

The fight that actually matters

Now the through-line.

Every choice above is, at root, a bet on where the agent runtime sits. Inside the CMS (Kontent.ai, Sitecore), the CMS vendor owns the runtime. Above the stack (Optimizely Opal), the suite vendor owns it. In your IDE or with your own LLM (Kentico KentiCopilot, Umbraco MCP), you own it, and the CMS becomes a tool your agent calls. With a different vendor entirely (Contentful into Salesforce Agentforce), the agentic platform vendor owns it, and the CMS becomes a content service.

That last one is the news of the year. A platform vendor that has built its entire 2024-to-2026 strategy around agents has moved to buy a CMS. Read the order of operations carefully. The agentic platform did not bolt on a CMS feature. The CMS got bought by an agentic platform.

Optimizely’s rebrand tells the same story from the other side. Its own announcement describes building for three audiences now: the marketer creating the work, the customer experiencing it, and the AI agents scraping, interpreting and acting on that content on the customer’s behalf. That third audience is the whole game, said out loud by the vendor.

Here’s the shift in plain terms. For 20 years the CMS was the library: a place you visited and pulled books off the shelf yourself. The agent layer changes the visit. Now an agent walks into the library on your behalf, fetches the book and summarises the relevant chapter. The library still has to be well-stocked and well-catalogued, because that’s where the work has gone. But customers have stopped walking through the door.

The question ‘which CMS?’ hasn’t gone away. It’s now sitting on top of a quieter and more important one.

Whose agent is walking through the door, and can you send your own?

Five questions worth asking

When a client briefs us on a platform selection in 2026, we run a separate pass on the agent layer. Five questions do most of the work.

  1. Where does the agent run? Inside the CMS, above the stack, in your own tools, or with someone else entirely? The answer tells you who holds the leverage in five years.
  2. Is the platform MCP-native, or do agents only work through proprietary plumbing? If you can’t point an external Claude, ChatGPT or Cursor at it, you’re locked in. MCP is the cleanest proxy for ‘leavability’ we’ve got right now.
  3. Are permission inheritance and approval gates explicit, or implied? Get them in writing. The agent that ships at 2am is the one nobody briefed you on.
  4. What does run quality look like? Scoring, drift detection, rollback, audit trail. Opal’s Quality tab is the current bar to benchmark against, because nobody else has shipped its match yet.
  5. What’s the override path when an agent ships the wrong thing? Who gets paged. Who can roll it back. How fast.

If a vendor can’t give you a clean answer to all five, the demo is doing more work than the platform.

What we tell clients now

Three things, in order.

First, AI inside your CMS is still not the most important selection criterion. The six pillars from our CMS selection guide (investment, capability, usability, security, scalability, reliability) still do most of the work. AI is now part of the floor, not the tie-breaker.

Second, the new criterion sits one floor up, and it’s the agent layer. Specifically, whether you can own it or at least leave it. MCP support, in production today or credibly on the roadmap, is the proxy we use. (If you want the argument for why model and platform portability is worth paying for, I made it here.)

Third, the platforms most likely to age well over the next five years are the ones that treat the CMS as a service the agent layer consumes, by design or by happy accident. Kontent.ai and Umbraco arrive there from very different directions. Optimizely is the one that has already shipped a runtime that does it. Salesforce will arrive by acquisition. The rest still have work to do.

Agentic CMS: Frequently asked questions

What is an agentic CMS? 

An agentic CMS is a content management system that can act on a goal, not just assist with a task. Where an AI-assisted CMS suggests a draft for you to approve, an agentic CMS takes an instruction like ‘localise this campaign for five markets’, plans the steps, uses the tools and systems it needs, and carries the work out, inside the guardrails and approvals you set.

How is an agentic CMS different from a CMS with AI features? 

AI features assist a human who stays in the driver’s seat: autocomplete, summaries, alt text on request. An agentic CMS can take initiative and complete multi-step work on its own. Optimizely put the difference well: a CMS that just drafts faster is no more agentic than a car with cruise control is autonomous.

What’s the difference between an agentic CMS and a headless CMS? 

They answer different questions. Headless is about delivery: content is decoupled from presentation and served to any front end through an API. Agentic is about action: software that can read your content, reason about a goal and do the work. The two pair well, because agents act far more reliably on structured, well-modelled content than on page-shaped blobs.

Which platforms are agentic CMS in 2026? 

Most major platforms now ship something agentic, but they bet differently. Kontent.ai and Sitecore put agents inside the CMS; Optimizely runs them above the suite; Kentico and Umbraco lean on MCP so agents can run in your own tools; and Contentful is set to become the content layer for Salesforce’s Agentforce. Which one is ‘agentic’ matters less than where the agent runs.

Who claims the ‘first agentic CMS’? 

Kontent.ai used the ‘world’s first agentic CMS’ line at its October 2025 launch. Optimizely had Opal agents in market earlier, and Sitecore launched SitecoreAI in November 2025. All three have a reasonable claim. The trophy matters less than the architecture underneath it.

Is an agentic CMS safe? Can AI publish without approval? 

An agentic CMS is only as safe as its guardrails. The 2026 floor is permission inheritance (agents act within a user’s existing rights), a posted ‘no publish without human approval’ policy, full audit trails, and a fast rollback path. Get these in writing during selection, not after an agent ships something at 2am.

What is MCP, and why does it matter for a CMS? 

MCP, the Model Context Protocol, is an open standard that lets external AI tools such as Claude, ChatGPT or Cursor connect to a system and act on it. For a CMS it’s the cleanest test of ‘leavability’: if you can point your own agent at the platform, you own the agent layer. If agents only work through proprietary plumbing, you’re locked in.

What’s the best agentic CMS to choose? 

There isn’t a single best one, and beware any demo that implies otherwise. AI is now part of the floor, not the tie-breaker, so the old criteria (investment, capability, usability, security, scalability, reliability) still do most of the work. The new question sits one floor up: can you own the agent layer, or at least leave it? MCP support is the proxy we use.

So what, plainly stated

The vendors are arguing about the word ‘agentic’. The practitioners are quietly making MCP the only thing that matters. The CMS is becoming a content service that the agent layer consumes, and the agent layer might belong to your CMS vendor, your suite vendor, your CRM vendor, or to you.

The right platform in mid-2026 isn’t the one with the most agents on its home page. It’s the one whose agent layer you can either own or leave.

Everything else is a pitch deck.


Sources and further reading

  • Optimizely rebrands as ‘the AI platform for marketing’ – Optimizely, 22 June 2026
  • What’s new in Optimizely Opal – Optimizely, 4 June 2026
  • AI-powered enterprise CMS: the impossible use cases – Optimizely (the ‘cruise control’ line)
  • Salesforce signs definitive agreement to acquire Contentful – Salesforce, 1 June 2026
  • Introducing Agentic Studio in SitecoreAI – Sitecore, 10 November 2025
  • Agentic CMS and Expert Agents – Kontent.ai / CMSWire, 31 March 2026
  • KentiCopilot and AIRA – Kentico
  • Umbraco product update, Q1 2026 – Umbraco (the MCP roadmap)
  • The Model Context Protocol – the open standard behind ‘leavability’
  • How AI is revolutionising the CMS – Luminary, the 2025 piece this follows up
  • How to choose the right CMS – Luminary, the six-pillar selection guide
  • WTF is GEO? and Optimizely vs Sitecore – Luminary, related reading

Want to tap into the expertise of an agency that’s been in operation since 1999?

Get in touch

Keep Reading

Want more? Here are some other blog posts you might be interested in.