Appendix E: Getting Started - From Current State to Agentic Workforce
Build Your Foundational Knowledge Agent (Your First Agent)
Foundational Knowledge Agent The first agent every VMS business should build. Purpose: centralize domain knowledge, deflect internal questions, and serve as the base layer for future agents.
This appendix provides an actionable roadmap for VMS businesses that have already begun their AI journey. It acknowledges that most businesses are not starting from zero and provides a framework for leveraging current initiatives to build a scalable, agentic workforce.
Step 1: Audit Your Current AI Usage (The First 30 Days)
Most VMS businesses have already deployed AI in some form. The first step is to conduct a comprehensive audit to understand your current state.
Common Current States:
General Use Tools: Employees are using public tools like ChatGPT or Microsoft Co-pilot for a variety of tasks.
Developer Tools: Your development team is using AI-enabled IDEs (e.g., GitHub Copilot) to accelerate coding.
Limited AI Agents: You may have deployed custom GPTs or other limited agents and are now facing scalability and integration challenges.
SaaS AI Add-Ons: You have upgraded existing SaaS products (e.g., Zendesk, Salesforce) with their native AI add-ons.
The Audit Process:
Survey Your Teams: Identify where and how they are using AI tools.
Collect Successful Prompts: The most valuable output of this audit is the collection of successful prompts that your teams have developed. A prompt that an employee uses repeatedly to get a good result is a proven, high-value use case waiting to be automated.
Map to Business Value: For each successful prompt, identify the business value it creates (e.g., time saved, errors reduced).
This audit provides the starting place for "agentifying" your processes.
Step 2: Deploy and Expand Your Foundational Knowledge Agent (Next 30 Days)
To prepare for an agentic workforce, we recommend starting with a strong Knowledge Base Agent. This agent is trained on the core knowledge of your business and products and serves as the foundation for many future agents.
Why start with a Knowledge Agent?
Baseline for Future Agents: Most agents you build will require this baseline knowledge to function effectively.
Immediate ROI: A well-trained knowledge agent can be immediately deployed for Tier 1 and Tier 2 support, providing a quick return on investment.
Extensible: By adding the codebase to its training data, this agent can be extended to offer Tier 3 support and other code-related capabilities.
The Deployment Process for Your First Agent:
Collect and Curate Training Data: This is the most critical step.
Connect key data sources (e.g., knowledge base, historical support tickets).
Collect and curate core documents (product guides, manuals, etc.).
Best Practice: Create a dedicated folder on SharePoint or a similar system to house all training data. Follow the data conversion process outlined in Part 7 to convert everything to AI-friendly formats (Markdown/JSON).
Outline Integrations and APIs:
Identify the external systems the agent needs to access.
Define the specific functions the agent will need (e.g., getCustomerDetails, createSupportTicket).
Create narrowly scoped API keys for each function to ensure a zero-trust environment where the agent is only authorized to perform its required tasks.
Test, Test, Test:
Use a platform with integrated tools for Human Feedback to refine the agent\\'s accuracy.
For conversational agents, use Human-in-the-Loop (HITL) capabilities to have a human approve responses before they are sent.
Perform end-to-end testing to ensure the data the agent is writing to and reading from third-party systems is accurate.
Go Live and Monitor:
Roll out the agent to a limited pilot group.
Continuously monitor its performance using HITL and human feedback to improve its efficacy over time.
Initial deployment should begin within 30 days; full stabilization should take less than 90 days for a single, well-defined agent. The complexity of the integrations may adjust this timeline, but the goal is rapid, iterative deployment.
By following this roadmap, you can leverage your existing AI experiments, build a strong foundation with a core knowledge agent, and create a scalable process for building out your agentic workforce, one agent at a time.
Last updated