Outline

1. Agentic AI Scope with an Example
2. How Current Agents Work and their Limitations
3. Brief introduction of Discoverable Destinations
4. Agents + Discoverable Destinations
5. Architecture & Sample Workflows
6. Discussion

Agentic AI Scope

• Autonomous systems that plan, reason, and execute tasks
• Minimal user input and automate multi-step workflows
• Understands natural language and acts on behalf of users

Agentic AI Example: Accessibility statement comparision

User asks: “Compare accessibility statements across 50 partner websites.”

• Ideal Agent Workflow:
• Discover relevant pages
• Navigate consistently
• Extract and summarize accessibility statements found

High Level Agent Workflow for Navigation

• HTML parsing
• Site-specific scripts

Limitations of Current Agents

• High maintenance
• Brittle
• No universal navigation standard

Discoverable Destinations

• Standardized semantic markers for common pages
• Examples: help, contact, log-in, accessibility-statement
• Enables predictable navigation

Stronger Together? (Agents and Discoverable Destinations)

• Discover destinations via semantic identifiers
• Navigate reliably to key pages
• Consistent automation across compliant sites

Stronger Together? (High Level Architecture & Workflows)

• LLM: Planning, reasoning, content synthesis
• Semantic Tools (based on Adapt's Discoverable Destinations): Discovery, navigation, content retrieval
• Complex Actions (like password changes): Combine destinations + APIs + human-in-the-loop

Sample Workflow & Tools

Tools, using Adapt Discoverable Destinations, could work as follows:

1. Retrieve list of discoverable destinations
2. Search for relevant destination
• Found? Navigate to destination
• Not found? Notify user, stop
3. Fetch content
4. Process, summarize and present to user

Discussion

Appendix

Optional Slides

Tool Integration using Model Context Protocol (MCP)

• MCP Overview: An open protocol enabling AI agents to securely connect with external tools and data sources.
• Roles:
  • MCP Clients: AI agents / LLM applications
  • MCP Servers: Tool providers (e.g., Semantic Web Tools)
• Workflow: Agents discover destinations → plan navigation or retrieve content → process with LLMs, if required.
• Note: MCP is one example, other protocols and frameworks may also be used.

MCP deployment models : How agents call these tools

• Traditional MCP Server (remote):
  • Runs tools on a server; great for public pages
  • Limits: no access to user’s browser session and no live DOM/UI control.
• WebMCP (in‑page):
  • The web page exposes tools to the agent via a JS API.
  • Strengths: access to full browser session and DOM control; ideal for authenticated, user‑specific flows.