Everybody Orchestrates. Most Do It by Hand. Photo by Giorgio Tomassetti on Unsplash Jürgen Haas Wed 16 Sep 2026 - 12:15

Every month I do my agency's billing by hand: export timesheets from one system, create invoices in another, merge, email, file, upload to the accountant. I maintain ECA. Elsewhere, a multi-agency project runs from an Excel sheet nobody can keep current, because the spreadsheet is the only place people see everything and feel in control. Everybody orchestrates, most by hand, and not for lack of tools. ECA, Maestro, FlowDrop, Tool API, AI Integration - ECA 1.0.0 and the Orchestration module with Activepieces, soon n8n, could run my billing end to end today. But the builder opens five UIs and, worse, has to decide which engine runs which step. No user can make that decision. The fix is one UI: every component of every participating system on one canvas, engine routing done by the platform. The Modeler API, the Workflow Modeler, Tool API's typed contract and Post 6's shared vocabulary are that architecture. Missing: a composite model owner, a dispatcher, the cross-system data contract. Let's build them. In Drupal.

Credentials still travel by email and by Slack on projects that are otherwise carefully built. Every integration with an external service needs them, and forwarding them from one party to the next is so routine that the risk rarely gets questioned. That habit is avoidable, and avoiding it changes who holds the secrets and who never has to see them.

The habit of sharing passwords over email or Slack

Receiving a password in plain text over email or Slack is a common occurrence on any Drupal project. It happens because every integration with an external service, whether a payment gateway, a CRM, or a third-party API, requires access credentials. Username and password pairs, tokens, private URLs, and other sensitive data that at some point need to travel from one place to another.

These credentials should never go into a code repository or into Drupal's YAML configuration files. The exposure risk is too high. The most common alternative, however, does not solve the underlying problem.

The burden of provisioning credentials in a project

The most widespread practice is to store secrets in a file outside the webroot, or to define them as environment variables accessible only to PHP. This works, but carries an operational cost that becomes apparent as soon as a password needs to change: those files must be edited by hand, and if environment variables are used, reloading Apache or Nginx may be required.

The deeper problem is the…

One rulebook for your AI: make every generated intro, summary and alt text follow the exact same governance as your chatbot.

Drupal 10 reaches end of life on 9 December 2026. Drupal 12 is planned for the same week. Our recommendation for Drupal 10 site owners is to start the Drupal 11 upgrade now instead of building the whole plan around a new release. Release week leaves no room for discovery As of 16 September, the official Drupal release schedule places Drupal 12 in the week of 7 December. Drupal 10 reaches end of life on 9 December. A planned release date does not promise that every module your site needs will be ready that day. Waiting until December to attempt an upgrade combines two problems: evaluating a new major version and dealing with the end of support for the existing one. Drupal 11 is already available for a trial upgrade. Our Drupal 10 end-of-life article covers the readiness checks. This…

Choosing between monolithic, decoupled, and hybrid Drupal should start with publishing cost - not a frontend framework preference.

Drupal architecture determines how many systems your team must maintain to deliver server-rendered HTML that readers and AI crawlers can use on the first response. Here is how to compare the three options by preview, metadata, cache invalidation, and operating work.

Joomla 3 has been unsupported since 2023. An honest look at the three destinations — Joomla 5, WordPress or Drupal — what actually breaks, and what it costs.

Join us THURSDAY, September 17 at 1pm ET / 10am PT, for our regularly scheduled call to chat about all things Drupal and nonprofits. (Convert to your local time zone.)

We don't have anything specific on the agenda this month, so we'll have plenty of time to discuss anything that's on our minds at the intersection of Drupal and nonprofits. Got something specific you want to talk about? Feel free to share ahead of time in our collaborative Google document at https://nten.org/drupal/notes!

All nonprofit Drupal devs and users, regardless of experience level, are always welcome on this call.

This free call is sponsored by NTEN.org and open to everyone.

Information on joining the meeting can be found in our collaborative Google document.

The safest way to add JSON-LD to Drupal is to map Schema.org properties to existing content fields with Metatag and Schema.org Metatag.

JSON-LD in Drupal should come from the same field model that supplies the visible page - not from hand-written scripts that drift when prices or availability change. Here is how to map tokens, export config, validate rendered pages, and catch missing bundles in CI.

Drupal Code Search now index recipes

Made it easier to get a list of projects that are included in a particular recipe. It's thanks to the sponsorship of Vardot, and previously Palantir.net that I'm able to spend time on tooling for the community. Many thanks to them.

Code search:

theodore September 15, 2026

Wondering why Drupal needs another page builder? Here's what I learned installing Drupal Canvas, exposing my SDCs, and writing React right in the browser.

Architecting a Headless RAG Engine with Drupal 11 bhavinhjoshi Tue, 09/15/2026 - 12:28

Drupal 11 is evolving into something far more powerful than a traditional Content Management System. For enterprise organizations, it is quickly becoming the foundational vector engine for secure, sovereign AI.

As organizations move beyond the hype of basic generative AI, the limitations of standard API wrappers become clear. Bolting a conversational UI onto a monolithic frontend, and blindly passing proprietary node data to public third-party models, introduces unpredictable latency, unmanageable token costs, and critical data compliance risks.

The Core Architectural Dilemma

When an organization attempts to integrate AI without a solid architectural foundation, the implementation typically fails through three specific avenues:

• The "Wrapper Module" Trap: Teams prioritize speed by installing pre-built chat widgets that pass unvetted, sensitive node data directly to external public APIs like OpenAI or Anthropic, compromising data sovereignty.
• The Fixed-Token Chunking Flaw: Scraping rendered HTML and relying on basic character-count chunking destroys semantic meaning, splitting context mid-sentence and returning poor vector matches.
• The SaaS Dependency: Relying on external, cloud-based vector databases creates a secondary point of failure and pulls proprietary organizational knowledge outside the compliant corporate network boundary.

The Headless RAG Solution

Many organizations view basic LLM integration as a complete AI strategy. They assume that passing a system prompt with full node text is sufficient for enterprise intelligence.

This is a mistake.

If an organization lacks the data governance to secure its AI pipeline, it will inevitably expose sensitive IP and face spiraling API costs. True digital sovereignty requires architecting native Retrieval-Augmented Generation (RAG) directly into your core infrastructure.

Engineering the Pipeline: A New Standard

To ensure the security and longevity of an enterprise AI implementation, organizations must shift from a "plugin" mindset to an "engineering-infrastructure" mindset:

• Vectorize the Entity API: Intercept entity events (nodes, taxonomy, media) at creation. Extract plain text from field data and attachments before the content is ever rendered to a frontend.
• Intelligent Ingestion & Chunking: Integrate Python and LangChain workers to handle semantic chunking strategies, ensuring extracted text is grouped into logical, context-rich units before vectorization.
• Native Vector Storage: Utilize PostgreSQL with the open-source pgvector extension to store high-dimensional embeddings natively alongside standard relational data, effectively air-gapping your intelligence layer.
• The Decoupled AI Endpoint: Expose the RAG pipeline as an authenticated, rate-limited HTTP endpoint (a JSON:API for AI) so any decoupled application can securely access grounded intelligence.

Final Thoughts

By treating AI not as a third-party plugin, but as core data architecture, Drupal 11 transitions from managing content to orchestrating enterprise intelligence. Stop treating AI as a shiny widget, and start building secure RAG infrastructure.

We don't believe in bolting on off-the-shelf wrappers. We believe in engineering systems that respect your investment and secure your data. If you are exploring enterprise AI, we approach architecture differently.

Drupal Drupal AI Drupal 11 Drupal Planet Share this Copied to clipboard Add new comment

Your AI chatbot is making up its own rules. Here's how to make every answer follow yours instead.

Today Acquia launched a new brand, and my favorite part is the updated logo. Right under the Acquia name, it now says "Powered by Drupal".

Drupal has always been at the core of Acquia, but for the past 5 years it was less visible in how we described ourselves. Now it's front and center again.

But that is not the main reason for the rebrand. The bigger reason for the rebrand is to help people see what Acquia has become. Our products have evolved faster than awareness of them.

The new brand leads with content instead of digital experiences, and the homepage calls Acquia an "agentic content platform" rather than a "digital experience platform".

Acquia Source is our new command center, bringing content management, digital asset management, and web governance into one workspace. Acquia AI coordinates agent work across those tools.

For agents to work safely across these tools, they need content they can trust and clear rules for using it. Somebody still has to decide what is approved, who can use it, and where it can go. I wrote about that in AI and the great CMS unbundling, and the new brand puts that idea at the center of our story.

Drupal is well suited for that job. Structured content, granular permissions, workflows, and revision history are the things agents need to work safely, and Drupal has refined them for years.

Customers can use Acquia Source CMS, our fully managed Drupal SaaS offering. For teams that want full control over their Drupal sites, we offer Acquia Cloud. Acquia Source brings sites on either platform into a shared workspace.

As you scroll the new homepage, it builds up our technology stack one layer at a time, starting from Drupal.

Acquia leaning into Drupal is also good news for Drupal itself. It helps close the gap between Drupal and its reputation. Drupal is still often seen as a CMS that requires a developer for everything, even as improvements in recent years have made it easier for marketers to build pages and manage content themselves.

So alongside the new brand, we'll be investing more in helping the Drupal community evangelize Drupal, reaching developers and marketing leaders who may not have looked at it in years. I want more people to see what Drupal has become.

Your next campaign launch does not have to stall on the landing page. You describe it the way you would brief an agency (audience, goal, story, call to action), and minutes later a production-ready page sits in your own Drupal site, built only from your organization's approved components, on brand and accessible from the first draft, ready to publish the same day.

Most articles with this title are written by Drupal shops. We build on both platforms, so here is the honest version: the four situations where Drupal genuinely wins, what actually moves versus what gets rebuilt, and why most WordPress sites should stay where they are.

Drupal's emerging agent interfaces are beginning to expose actions as well as information. MCPIO 1.0.0-alpha1 presents external Model Context Protocol clients with two entry points, search and execute, through which they can discover and invoke a wider set of Drupal operations. WebMCP User Forms takes a different route, adding WebMCP-style tool metadata to Drupal's login, password-reset, and registration forms so browser agents can identify their inputs more explicitly. These are different interfaces, but both make capabilities that were designed primarily for people or Drupal code more legible to automated actors.

Once a capability becomes callable, the harder question is not simply whether an agent can find it. Will Huggins' argument about Drupal publishing capabilities puts authentication, permissions, workflow, guardrails, and controlled context alongside the interface itself. In that model, an agent-facing tool should not create a route around Drupal's existing governance controls; the authority to perform an action still has to come from an identified actor with defined access.

The current projects do not establish one settled architecture for doing this. MCPIO remains an alpha release and is not covered by Drupal's security advisory policy, while WebMCP User Forms is also an alpha project outside that coverage. MCP, WebMCP, Tool API, and other agent-facing approaches also operate at different layers, from external protocol clients to capabilities exposed inside a browser. That distinction matters because making an operation discoverable is not the same as deciding who may execute it.

Issue 36 of Editor's Pick looked at a broader requirement for machine-facing websites: meaning must be explicit, access constrained, and important rules represented in ways software does not have to infer from prose. The next question is narrower. As Drupal makes more capabilities available to agents, interface design and permission design begin to meet at the same boundary: which actor is represented, which actions that actor may invoke, and which Drupal controls remain in force when the request comes from software rather than a person. The implementations are still emerging, but that authority boundary is becoming harder to treat as a secondary concern.

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.

Generating software is becoming easier. Knowing what went into it, whether it can be trusted, and who will maintain it is becoming the harder problem.

Using AI to review AI-Generated code: Astra vs. Opus vs. DeepSeek John Locke Sun, 09/13/2026 - 19:00 Astra built an impressive amount of working code, but missed serious performance problems. DeepSeek found several of them for 22 cents. Opus cost dramatically more and found a substantially different set of problems, especially around UI and accessibility. Dev Corner

• Read more about Using AI to review AI-Generated code: Astra vs. Opus vs. DeepSeek
• Add new comment

Handing an outside developer your one and only SSH login because the job was one theme on one site is the oldest mistake in hosting, and the collaborator button on the big platforms only hides the question of what that seat can actually do. On a BOA server a Client, the control panel's word for the person a site belongs to, gets its own limited login built automatically, one which sees that Client's sites and their files and nothing else; and now your host can turn on a second login of the platform kind for that Client, which reaches the whole codebase of every platform whose sites are all the Client's, with Composer and the Drush shipped inside the platform, and refuses any platform where somebody else's site is parked as well. The grant is decided by root from one line, re-derived every three minutes and readable in one file and one log; Drupal 8 and newer platforms are unlocked for the developer an hour at a time, on request, and locked back by themselves.