AI assistants increasingly sit between people and websites, retrieving and interpreting information on their behalf. That shifts the problem beyond whether software can connect to a site at all. The system also has to make clear what information means, what software may access, and which rules govern what happens next.
The Drupal AI Initiative made the first part of that problem explicit on 24 August 2026. Fields, taxonomy, relationships, metadata, and other forms of structured content can give software information that is easier to identify and interpret than meaning conveyed mainly through page design. Permissions and workflows add another layer by distinguishing what is public, restricted, current, or subject to editorial control.
A more concrete access boundary appeared this week in Acquia DAM's MCP Server. Compatible AI assistants can search assets, retrieve metadata, and obtain links through an authenticated connection that follows the user's existing DAM permissions. The initial implementation is read-only: an assistant can retrieve permitted information but cannot edit metadata or upload files. The interface therefore exposes capability without giving the software unrestricted authority over the system behind it.
GovCMS addresses a different machine-readable problem. On 27 August 2026, Rules as Code became a production capability on the platform, allowing policy, legislation, and business rules to be represented as reusable digital assets and embedded in digital services. Its first production implementation is an eligibility checker for agencies considering GovCMS, and the underlying eligibility rules are also published as open source. GovCMS describes the approach as a foundation for future AI-assisted services rather than as an AI-agent implementation today.
These developments are not three versions of the same architecture. They expose three related requirements for systems that software increasingly has to interpret or use: meaning must be explicit, access must be constrained, and important rules should not have to be guessed from prose. For Drupal teams, the agentic web may therefore depend as much on long-standing work around content modelling, permissions, auditability, and governance as it does on new AI capabilities.
Follow The DropTimes on LinkedIn, X, Bluesky, and Facebook, or join #thedroptimes on Drupal Slack.
This issue of Editor’s Pick was written and curated by Allen Jason.
