From Chatbots to AI Agents in Drupal

Government websites have traditionally been designed around navigation. Visitors use menu systems, search boxes, and links to find and browse content. Chatbots made the experience more conversational, but frequently, are just an alternative way to find content. Agentic solutions, called AI agents, on the other hand, help users accomplish goals and tasks.

With an agent architecture you can define multiple agents, each an expert in a particular domain. For example, a visitor might ask, “I need to renew my fishing license and find out when trout season begins.” An orchestrator agent can route the request to specialized licensing and regulations agents, retrieve information from trusted sources, and return one coordinated response. This marks a shift from websites that simply publish information to intelligent digital services that understand user intent and help people accomplish meaningful tasks.

In this article, we’ll look at how Drupal’s AI and AI Agents modules can power the shift to agentic architecture by complementing existing navigation based websites. We’ll also look at why Drupal is an ideal platform for enterprise agentic experiences that are grounded in your organization’s structured content, APIs, and workflows.

Why agentic experiences are the next step

Traditional websites require users to navigate menus, search results, PDFs, and forms, and terminology that may reflect the structure of the organization rather than the needs of the visitor.

Agentic experiences offer a different model – instead of requiring users to determine where information lives, the agentic approach powers modern conversational interfaces that allow users to simply describe what they need:

“When does deer season start in my county?”

“Which court form do I need for a small claims case?”

“What jobs match my background in IT and project management?”

AI agents can interpret the user’s intent, retrieve information, perform searches, interact with internal and external systems, and return a direct, contextual response. Additionally, agents can move beyond answering individual questions to helping users accomplish multi-step tasks.

For example, a court assistant might leverage agents that identify the appropriate form, explain the filing requirements, provide the filing fee, and direct the user to the correct submission process. A licensing assistant could utilize agents to verify eligibility, retrieve current regulations, and guide the user to the appropriate application.

This is a key difference between an informational chatbot and an agentic experience: the chatbot provides the conversational interface, while agents perform the specialized work required to fulfill the user’s request.

Users no longer need to understand how an organization divides its content, departments, and digital services. The agent layer can coordinate and compile those resources on demand. This shifts government websites from being information repositories that users must navigate into intelligent assistants helping people find answers, make decisions, and accomplish tasks.

More differences between chatbots and AI agents

The terms chatbot and agent are sometimes used interchangeably, but within the Drupal AI ecosystem they describe distinct layers of the application.

Chatbots provide the conversational interface. In Drupal, a chatbot is the user-facing interaction layer. It displays the ongoing conversation, accepts the user’s questions, and presents the resulting answers, links, forms, or other structured responses. The chatbot does not perform the underlying work itself. Instead, it passes the user’s request to an AI assistant, an AI agent, or directly to a tool such as Semantic Search.

AI agents perform specialized tasks. An agent is the task-performing layer. It uses an LLM to interpret the user’s request (passed from the chatbot), determines what actions are required, invokes the appropriate tools, and returns the results to the chatbot. Depending on its task or purpose, an agent might use tools to:

• Query Drupal content and entities
• Perform traditional or semantic (similarity) search
• Retrieve or create information through an internal or external API
• Initiate a form, workflow, or other business process
• Call custom Drupal services and functionality

For example, a Hunting Regulations agent might search approved Drupal content for the regulations associated with a particular species, season, and location.

Separating the chatbot from the agents makes the architecture easier to maintain. The conversational interface can evolve independently, while the specialized agents can be reused, assigned permissions, and connected to the tools, content, and systems they need.

Orchestrating agents

A single agent may be sufficient for a focused assistant, such as a semantic search assistant or an FAQ chatbot. More complex requests, however, may span multiple topics, domains, or steps.

Agentic architectures can support more complex requirements and experiences by coordinating multiple specialized agents, each with its own responsibilities, tools, permissions, and sources of information.

Within Drupal, an orchestrator agent can serve as the coordinating layer between the chatbot and one or more specialized agents. It interprets the user’s overall request and determines which agents are needed, delegates the appropriate work, and combines the results into a single, coherent response. Depending on the request, the orchestrator agent may:

• Route a request to the agent best suited to handle it
• Divide a multi-part request among several agents
• Pass relevant information (context) and results between agents
• Coordinate the results into a unified response for the chatbot

The orchestrator agent does not need to contain all of the underlying domain knowledge itself. Instead, it just identifies the appropriate, available agent that can accomplish the given tasks.

Example of an agent workflow. A user asks: “I need to renew my fishing license and want to know when trout season starts.” The orchestrator agent identifies that the request contains two related but distinct needs, sends the license-renewal portion to the Licensing Agent, sends the seasonal-regulations portion to the Seasons Agent, receives the renewal requirements, application link, and applicable trout-season dates, and combines the results into one coordinated response for the chatbot to present. From the user’s perspective, this remains a single conversation with the assistant. Behind the scenes, the orchestrator coordinates the specialized agents and systems needed to fulfill the request.

Fallback agents and human escalation

Not every request can, or should, be handled entirely by the chatbot or AI. A well-designed agent architecture includes a fallback agent that can gracefully respond when a request falls outside the system’s supported capabilities or requires assistance from an authorized staff member.

Within an orchestrated experience, the orchestrator can route a request to the fallback agent when:

• No specialized agent is appropriate for the request
• The required information cannot be found in an approved source
• The request requires human judgement, review, or authorization
• The user appears frustrated or asks to speak with a person
• The request involves a sensitive, urgent, or high-risk situation
• The agent has low confidence on how to respond

Depending on the use case, the fallback agent might ask a clarifying question, direct the user to other trusted resources, open a contact form, create a support ticket, or initiate live chat. When escalation occurs, the system can also pass along the conversation history and relevant details already provided by the user, creating a smoother handoff and preventing the user from having to start over.

Profile data and personalization

Existing user data can help agents provide more relevant assistance throughout the conversation. When a user is signed in, an agent can use selected account and profile information to better understand the user’s circumstances — their location, communication preferences, existing licenses and applications, education and experience, or other types of information they provided through earlier interactions.

Personalization can reduce the need for users to repeat information and allow the assistant to provide more useful, personalized guidance. A career coach could tailor occupation recommendations to the user’s experience, education, and goals. A hunting/fishing regulation assistant could tailor information based on past requests and location preferences.

Personalization should be intentional and limited to the information needed for the task. Organizations can control which agents have access to profile data, and should clearly communicate how that data will be used. It is important to note that authentication and personalization are not required — the same chatbot can support anonymous visitors while providing additional, account-specific assistance to authenticated users.

Conversational analytics

Conversational analytics provide a level of observability that is not available in traditional click-based analytics. You can view the full questions users are asking, what answers the agents are providing, and follow the full thread of each conversation.

Traditional analytics tell you what pages users visited, what links they clicked, and where they entered and exited the site. Conversational analytics instead reveal what users are actually trying to accomplish, information they cannot find (content gaps), terminology users rely on that differs from the organization’s own language, requests that frequently result in fallback or escalation, and emerging trends in topics and services over time.

Conversational analytics can inform updates to Drupal content, search indexes, agent instructions, and workflows, creating an ongoing feedback loop. To address privacy concerns, conversational analytics can be completely anonymized or aggregated where appropriate.

Examples of agent-driven chatbots

Regulations and Seasons Assistant — a hunting regulations assistant for a Department of Conservation, comprised of a Regulations agent, a Seasons agent, and a Fallback agent. Example queries: “When does deer hunting season start this year?” or “What dates can I hunt with my crossbow?”

Court Assistant — helps self-represented litigants find information, court forms, and general help, with agents for Court Help, Court Forms, FAQs, and Search. Example queries: “How do I file a small claims case?” or “Which form do I use to request a restraining order?”

Career Coach — an assistant that acts as a career coach, combining Intake, O*NET occupation data, Resume helper, Interview helper, Occupations, Job Search, FAQ, and Fallback agents. Example queries: “What careers match my interests?” or “What career paths are available after military service?”

Staff Help Desk — a private chatbot for content creators and Drupal administrators, with Content Type, Component, and Style Guide agents. Example queries: “What component should I use to display photos from our last event?”

Content Governance Assistant — a private chatbot with Content freshness, Accessibility, SEO, and Brand compliance agents. Example queries: “Which pages have not been updated in 3 years?” or “List all pages that have broken links.”

Future-proofing with an agentic architecture in Drupal

The web is evolving from navigation-driven experiences toward conversational and agent-driven experiences. Users increasingly expect to ask questions without having to understand how an organization structures its departments, content, and systems, receive direct and contextually relevant answers, and complete tasks through conversation.

Organizations that already maintain content, workflows, and business processes in Drupal are well-positioned to take advantage of this shift. Drupal can provide the foundation, while specialized agents make that information and functionality available through conversational experiences. An agentic architecture allows organizations to evolve incrementally — they might begin with a focused FAQ, search, or forms assistant and later introduce orchestration, additional domain agents, personalization, workflow integrations, analytics, and human escalation.

Because the chatbot, agents, and tools remain separate layers, each can be refined or replaced as organizational needs and AI technologies change. The same underlying architecture can support many different experiences, from public-facing licensing, career coaches, and court assistants to internal help desks and content governance tools, while it leverages existing Drupal content, APIs, permissions, and governance practices used for traditional websites.

By combining these capabilities, Drupal becomes more than a content management system. It becomes a flexible platform for intelligent digital services that help people find answers, make decisions, and accomplish tasks.

Learn more

• Find out more about using AI to optimize your organization’s digital goals
• Contact us about adding AI agents to your site