AI Agents for Content Editors

Hackathon Demo Walkthrough,

Drupal AI Hackathon - Play to Impact: 2026 edition

Context: The Challenge We Accepted

At the Drupal AI Hackathon (Play to Impact 2026), we participated in

Challenge #1: AI Agents for Content Editors.

The core question was simple but powerful:

Can AI make life easier for content editors?

The challenge emphasized three key focus areas:

• Automation and workflow orchestration
• Data ethics and governance
• Human-in-the-loop collaboration

Rather than treating AI as a content creator, we focused on AI as a supporting assistant — helping editors validate, optimise, and maintain content quality at scale.

---

Our Approach (High Level)

Instead of using traditional ECA (Event-Condition-Action), we built our solution using FlowDrop, a visual workflow orchestration tool we are very comfortable with.

This allowed us to:

• Design AI-powered editorial workflows visually
• Orchestrate on-demand and scheduled validations
• Ensure nothing is written automatically without human approval

The result is a Drupal site that behaves like an intelligent content hub, working with editors — not instead of them.

---

Scenario A – On-Demand Content Validation

Target audience: Business & Content Teams

The Problem

Editors often need quick feedback:

• Is my content SEO-ready?
• Are metadata fields missing or weak?
• Does this content meet quality standards?

Doing this manually is slow and inconsistent.

The Solution

We added a custom “Validate” action directly to Drupal’s content listing.

From the editor’s perspective:

1. Select a node
2. Click Validate
3. Receive AI-generated suggestions

Figure 1: Content listing with custom Validate action

Figure 2: Select content and trigger validation. User is prompted to either accept/reject validation.

Figure 3: Review AI-generated suggestions before applying

What Happens Behind the Scenes

• The node’s content is sent to an AI validation agent (using Mistral AI, as required by the hackathon)
• The agent checks multiple fields (e.g. title, description, SEO metadata)
• If improvements are needed, suggestions are generated

Human-in-the-Loop by Design

Crucially:

• The content is not updated automatically
• The editor must explicitly accept or reject the suggestions

This ensures:

• Editorial control
• Trust in AI output
• Compliance with governance and ethics requirements

AI assists. Humans decide.

---

Scenario B – Scheduled Validation at Scale

Target audience: Business & Editorial Leads

The Problem

On-demand validation is useful — but editors shouldn’t have to:

• Wait for AI responses
• Retry validations multiple times
• Manually check hundreds of pages

The Solution

We introduced scheduled AI validations running via cron.

Figure 4: Configuring scheduled validations via cron

Figure 5: Automated background scanning in progress

Figure 6: Dashboard showing content flagged for editorial review

How it works:

• All content is periodically scanned in the background
• AI detects potential issues (SEO, fact consistency, metadata quality)
• Violations are reported — not automatically fixed

Editorial Dashboard

Editors get a dedicated dashboard showing:

• Which content needs attention
• What type of issue was detected (e.g. fact check, SEO)
• A clear to-do list

Again:

• No content is changed without human approval
• Editors stay in control, but AI does the heavy lifting

---

Scenario C – Visual Workflow Orchestration

Target audience: Site Builders & Developers

This is where FlowDrop shines.

Workflow-Driven AI

All logic is built as visual workflows, for example:

• Fact checking workflows
• SEO metadata generation
• Quality scoring
• Scheduled batch processing

Example: Scheduled Validator Workflow

• Triggered by cron (every minute for demo purposes)
• Loads batches of content
• Processes each node in a loop
• Executes multiple AI validations per node

Each validation:

• Converts node data into structured prompts
• Requests JSON-based AI output
• Creates validation entities (to-do items)
• Waits for human confirmation before applying fixes

This makes the system:

• Modular
• Extensible
• Easy to reason about visually

Figure 7: FlowDrop visual workflow canvas

Figure 8: Cron-triggered batch validator workflow

Figure 9: Configuring AI validation nodes

Figure 10: Loop logic for processing content batches

Figure 11: Complete workflow with human-in-the-loop approval

---

What We Built During the Hackathon

Target audience: Developers & Contributors

Not everything made it into the final demo — but the experimentation mattered.

Key Outcomes

• Asynchronous human-in-the-loop support

  • Interrupt workflows
  • Resume later
  • Allow approval by different users, at different times

• FlowDrop AI Agents

  • Use Drupal AI agents directly inside workflows
  • Developed by my colleague David during the hackathon

• Workflow-to-Workflow Calls

  • One workflow can trigger another
  • Enables clean orchestration and reuse

• SEO Dashboard (Experimental)

  • Custom-built during the hackathon
  • Not open-sourced yet
  • Can be contributed if there’s community interest

Open Source vs Business Logic

• Generic tooling → open-source candidates
• Business-specific workflows → configuration, not code

This keeps things:

• Flexible
• Maintainable
• Practical for real-world use

---

Key Takeaways

• AI is most powerful when it supports, not replaces, editors
• Human-in-the-loop is not optional — it’s essential
• Visual workflow orchestration makes AI understandable and governable
• Drupal can evolve into an intelligent editorial platform, not just a CMS

Drupal