BRAINSUM reaches Gold: What contributing to Drupal AI taught us
BRAINSUM has reached the Gold tier of the Drupal Certified Partner program. For teams managing complex websites, our contributions demonstrate the expertise needed to reduce repetitive editorial work, keep AI under human control, and preserve the freedom to change providers.
A deliberate investment in Drupal AI
Since February 2026, we have dedicated significant resources to the Drupal AI Initiative and related projects. We made it because AI was fast becoming part of what it means to build a modern CMS.
Our public Drupal.org contribution record currently shows 32 issue credits.
Two-thirds of those credits are in AI-related projects:
- Context Control Center (CCC)
- AI (Artificial Intelligence)
- Drupal AI Initiative
- AI Agents Views
- AI Agents Test
- AI Content Review
- Anthropic Provider
- AI Image Classification
The record also includes practical initiative work such as recipes for AI-assisted comment-spam scoring, automatic unpublishing and content pre-moderation, alongside contributions to Drupal core and other modules. These are not isolated experiments. They sit close to the capabilities our clients are asking us to design, integrate and operate.
Was it worth it?
Yes, but not primarily because Bronze became Gold.
The strongest return was compressed learning. Working in the initiative exposed us to architectural discussions, integration edge cases and quality expectations before they reached a polished release or a client backlog. Contribution forced us to understand not only how a module works when the demo succeeds, but how it behaves across providers, content models, permissions, editorial states and failure modes.
That depth changes the questions we ask at the start of an engagement. Instead of beginning with ‘Which model should we use?’, we can begin with the business decision that needs to improve, the content and permissions the model may access, the human checkpoint required and the evidence that will show whether the feature is actually useful.
What we learned
1. The best first AI use cases are narrow, frequent and measurable
A general-purpose chatbot attracts attention, but focused editorial tasks often create value faster. Summarising a long article, proposing alternative text, translating a structured page, applying taxonomy or classifying an incoming message all have a clear input, a reviewable output and an identifiable person who benefits.
A useful first step is to choose one repetitive editorial task, agree how to measure time saved and output quality, and test it with the people who will use it. Their feedback should guide the next iteration. On VisitEurope.com, AI supports one-click translation, alternative-text generation, metadata creation and content enrichment inside Drupal. These features remove repetitive work while keeping editorial control visible. The project has now been nominated for the 2026 International Splash Awards in the Government & Public Services category. At DrupalCon Rotterdam, Krisztián Kása and Zsófia Alföldi will present One Brand Voice from 35 Sources, explaining how content strategy, Drupal architecture and generative AI turned 35 distinct national voices into a governed editorial platform without removing human control.
2. Context matters more than a slightly better model
Model benchmarks change quickly. The durable asset is the context around the model: approved terminology, brand voice, accessibility rules, regulatory constraints, content scope and examples of acceptable output.
Our work on the Context Control Center reinforced this lesson. Context should be managed like content, with permissions, revisions, moderation, multilingual support and clear scope. That is a natural fit for Drupal. It also makes AI behaviour easier to govern than a prompt copied into a third-party dashboard and forgotten.
3. Drupal’s content architecture becomes an AI governance layer
Drupal already knows who may see or change content, which revision is published, how translations relate, what workflow state an item occupies and how structured fields are validated. A responsible AI integration should respect those rules rather than bypass them.
This is a significant advantage for enterprise and institutional use cases. AI output can enter an existing moderation workflow, inherit access controls and remain connected to the source content and its revision history. The CMS becomes the controlled operating environment for AI, not just a source of text sent to an external service.
4. Provider flexibility is operational risk management
AI providers differ in capability, cost, latency, hosting geography, privacy terms and release behaviour. They also change APIs and model defaults. Our contribution to the Anthropic provider included resolving an incompatibility around model parameters: a small example of the integration details that determine whether a feature is reliable in production.
Drupal AI’s provider architecture lets a solution choose among commercial services, cloud platforms and self-hosted models. That flexibility gives your organisation more control over its technology choices. Changing providers still requires retesting prompts and behaviour, but a shared integration layer reduces the work tied to any one supplier and makes future changes more manageable.
5. Human review, testing and observability are product features
AI can produce a fluent answer that is wrong, inconsistent or inappropriate for the context. Production readiness therefore requires more than a good prompt. It requires defined thresholds, logs, cost visibility, evaluation examples, fallbacks and a clear owner for the result.
Our principle is straightforward: people remain accountable. For a low-risk task such as suggesting metadata, a quick editorial review may be enough. For automated moderation, publishing, accessibility or incident classification, controls must be stronger. The degree of automation should follow the consequence of an error.
6. Contribution is one of the fastest routes from experimentation to expertise
Using a module teaches you its interface. Contributing teaches you its assumptions, boundaries and roadmap. Reviewing issues from other organisations also reveals recurring requirements that no single client brief will show.
That knowledge moves in both directions. Client projects give us real operational constraints; community work lets us improve reusable solutions rather than patch the same problem privately on every site. The result is better engineering for clients and stronger software for everyone.
From experiments to a Drupal AI delivery capability
Our Drupal AI work now spans several layers of the digital platform:
- Editorial assistance: summaries, metadata, alternative text, translation and content enrichment.
- Discovery: semantic search and retrieval-augmented assistants that answer from selected Drupal content.
- Accessibility and formats: text-to-speech and structured content transformation.
- Automation: tagging, classification, moderation and agentic workflows for operational signals.
- AI discoverability: structured data, curated Markdown and llms.txt strategies for organisations that want their content understood by answer engines.
- Architecture and governance: provider selection, access boundaries, context management, evaluation and human review.
Back in November 2024, our Drupal RAG chatbot demonstration already showed how selected website content could power semantic retrieval while preserving a choice of language model and vector database. Since then, our focus has expanded from demos to the less glamorous, but more valuable work required for production: governance, editor experience, resilience and measurable outcomes.
What Gold means for organisations selecting a Drupal partner
Certification is not a substitute for due diligence. Ask prospective partners for evidence of delivery against real deadlines, experience with your integrations and a demonstration of the workflows your editors will use. Check architecture, security and accessibility, then agree who owns delivery, how the team will maintain continuity and what support and service levels apply after launch. But the Drupal Certified Partner tier adds an independently verifiable signal: the agency invests in the platform it proposes to clients.
For organisations evaluating an open-source CMS or exploring how AI can improve content operations, our Gold status points to something concrete: BRAINSUM combines long-term Drupal engineering experience with hands-on knowledge of the Drupal AI ecosystem. We can help define the right use case, select an appropriate architecture, integrate it into real editorial and governance workflows, and remain accountable after launch.
The badge is a milestone. The capability - and the responsibility that comes with it - is the real result.
Considering Drupal or Drupal AI for your next platform?
If your organisation is preparing a vendor shortlist, modernising a Drupal estate or deciding where AI can create measurable value without surrendering control of your content, talk to BRAINSUM. Bring us one editorial workflow you want to improve, the systems it needs to connect to and the deadline you are working towards. We can help define a focused first release, agree how to measure its value and plan the support it will need after launch.
