Blog

TYPO3 Is Great for Human Editors. It Is Not Built for AI Agents.

20+ years building TYPO3 projects. Real, Volkswagen, Focus, Immowelt, RWTH Aachen. Some of them enormous, some of them running for a decade with a hundred editors working in parallel.

Here is my honest conclusion after all of it: TYPO3 is a great CMS for human editors, and it is not a great CMS for AI integrations. The problem is not the idea, it is the architecture. Themes, content elements, and caching are too rigid and far too complex to hand to an AI agent.

Where TYPO3 still wins

TYPO3 does what it was designed to do extremely well. It runs large editorial teams, deep content trees, and heavy multilingual sites without falling over, and it has never let us down. If your business means human editors editing by hand, TYPO3 remains a solid, reliable choice.

Where it breaks the moment agents show up

AI agents need three things: a codebase they can read consistently, a predictable way to create structure, and no hidden state.

In TYPO3, a single small content element means a pile of files across nested folders, TypoScript, FlexForms, TCA, and registration files. Themes are just as rigid. Changing a style usually means touching configuration and templates rather than writing CSS. And caching is invisible state: you clear caches and hope the page does not return a 500 error.

None of that is a defect for human teams. It is fatal for agents. The documentation and the core are not consistent enough for an agent to learn the logic once and then apply it reliably, which is why my agents produced constant errors and why I stopped trying to use TYPO3 with AI.

What a modern CMS has to deliver natively

  • An API for website and content management, including SEO

  • Real multilingual content instead of a pile of translation plugins

  • Machine-readable content output, so the site can be consumed headless

  • Frontend user management for portals and gated content

  • OAuth login

  • A codebase that AI agents and humans can both read and understand

And one capability that separates the platforms: an editor describes the page they want, and the system writes the copy, then generates the content element and the matching CSS style on demand. Nobody will hand-build templates for years to come.

KocoDesk: the alternative I wanted to work with

I built KocoDesk around exactly those requirements, so I know how far the gap actually is. It is the multilingual CMS my AI agents operate through MCP, while human editors run the same core in a visual backend.

Your own agents, Claude, ChatGPT, or tools your team builds, connect through Model Context Protocol and build the page tree, create pages and navigation, write and localize copy, and publish to any market. Every one of those actions runs inside the same role-based permissions, approval gates, and audit trail as a human editor, and nothing goes live without human confirmation.

Your editors stay in control in the visual backend, and the platform handles what TYPO3 projects usually spend weeks on:

  • One core for every site and language, with native language catalogs, separate domains and themes per market, /en/ and /de/ routing, language fallbacks, and localized SEO

  • Every public page publishes as clean Markdown, and each site serves an /llms.txt file, so ChatGPT, Perplexity, and Claude can find, read, and cite your content accurately

  • Built-in frontend authentication, gated content, Google sign-in, and an OAuth2 identity provider, so customers use one login across your portals, tools, and CRM

  • A native 12-column grid with an inline visual editor, so layouts hold up everywhere without touching code

  • Fully managed delivery: hosting, updates, and security are on us with a fixed monthly price, not per site or per user, and migration of your existing site is part of onboarding

The conclusion I keep coming back to

TYPO3 will keep working well for human teams, and it deserves the reputation it has. But the next five years of publishing will run on agent-native stacks, and retrofitting that onto a core built before AI agents existed is not a project I would sell to a client.

That is the entire reason KocoDesk exists. If your CMS is fighting your AI strategy instead of supporting it, book a free consultation and we will look at your sites, your languages, and your rollout:

Book a Free Consultation