Martin Anderson-Clutz at Acquia recently made the case that Drupal is becoming something more specific than a CMS with AI features.

A governed AI harness.

The argument is that Drupal already has most of what agents need:

• Structured content
• Granular permissions
• Editorial workflows
• Revision history
• Centralized governance
• Typed fields and entity relationships
• Canvas AI and the Context Control Center

The funny part is that many of these are the same things that historically made Drupal feel cumbersome to human editors.

The platform famous for asking people to complete seventeen fields may have accidentally spent two decades preparing for robots.

That is legitimately interesting to me.

If an agent is working across hundreds of pages, multiple brands, regulated workflows, and several languages, complexity may stop being baggage.

It may become the safety system.

I covered the argument in the latest episode of Headless:

YouTube
Spotify
Apple Podcasts

Drupal Association article for anyone who wants the source directly:

https://www.drupal.org/about/ai/initiatives/blog/from-headless-cms-to-ai-harness-what-i-took-to-decoupled-days

For people actively building with Drupal AI, does the “AI harness” framing match the reality yet?

Or is the positioning currently ahead of the product?

Martin Anderson-Clutz at Acquia recently made the case that Drupal is becoming something more specific than a CMS with AI features.

A governed AI harness.

The argument is that Drupal already has most of what agents need:

• Structured content
• Granular permissions
• Editorial workflows
• Revision history
• Centralized governance
• Typed fields and entity relationships
• Canvas AI and the Context Control Center

The funny part is that many of these are the same things that historically made Drupal feel cumbersome to human editors.

The platform famous for asking people to complete seventeen fields may have accidentally spent two decades preparing for robots.

That is legitimately interesting to me.

If an agent is working across hundreds of pages, multiple brands, regulated workflows, and several languages, complexity may stop being baggage.

It may become the safety system.

I covered the argument in the latest episode of Headless:

YouTube
Spotify
Apple Podcasts

Drupal Association article for anyone who wants the source directly:

https://www.drupal.org/about/ai/initiatives/blog/from-headless-cms-to-ai-harness-what-i-took-to-decoupled-days

For people actively building with Drupal AI, does the “AI harness” framing match the reality yet?

Or is the positioning currently ahead of the product?

Cloudflare recently introduced EmDash, an open-source CMS designed around AI agents.

Its premise is that traditional CMS platforms were built for humans:

Humans open dashboards.

Humans install plugins.

Humans edit pages.

Humans hit publish and immediately notice the typo.

An AI-native CMS starts with a different assumption. The builder, administrator, or primary user might be an agent interacting through an MCP server or CLI.

That does not make EmDash a WordPress killer. WordPress has an ecosystem that cannot be recreated by announcing a GitHub repository and believing in yourself very hard.

But I think the architectural question is real.

Can WordPress become truly agent-native through plugins and new APIs?

Or will AI-native content management require systems designed from scratch around permissions, rollback, structured context, and machine-readable operations?

I covered EmDash, WordPress, Drupal, Contentful, and the wider CMS identity crisis in Episode 2 of Headless:

YouTube
Spotify
Apple Podcasts

I’m especially curious what WordPress developers think.

Renovate the house?

Or admit it was built before electricity and start drawing new blueprints?