Agentic Reasoning Engine

The Agentic Reasoning Engine is an LLM-powered reasoning architecture.

The Reasoning Engine (1) plans multiple steps to accomplish a goal, (2) select & execute the next action, and then (3) observes the results & adapts the plan.

The agentic architecture that powers the Moveworks AI Assistant is purpose-built to handle the unique challenges of supporting employees at enterprises across the world in a scalable, autonomous way. Every element of this architecture plays a critical role in ensuring that the AI Assistant is able to answer any question that a user might need help with.

Why true autonomous employee support needs an agentic architecture to deliver impact

In a nutshell, we have optimized for the following fundamental problems:

1. Employees need both information and the ability to take action on that information to get things done
2. Employees often phrase their needs in underspecified or ambiguous ways
3. Employees may not know how their problem may be solved, or where the solution to their problem lies

This architecture addresses a set of critical requirements that must be met for a performant, autonomous system:

1. The range of possible combinations of requests and solutions is huge, so it needs to be able to iteratively explore and refine the choices since each request can be different - Think of all the tasks that an enterprise-wide employee support system needs to handle across IT, HR, Sales, Procurement, and many more
2. The possible solutions are highly customized and are heterogeneous across organizations and even within organizations - The ideal system is tailored to the needs of your organization
3. The range of solutions and use cases should grow over time without being bound by the architecture - This system grows and evolves over time to offer employees and stakeholders more productivity and time savings over time with more diverse use cases
4. Organizations require a high degree of control and precision - It’s critical to have the ability to adapt and mold reasoning for solutions and resources to make sure that users get the right answers, even if the system can’t get it 100% right out of the box

Key Moveworks AI Assistant agentic components

The AI Assistant has the following key components which are essential to meeting these requirements:

1. A central LLM-powered reasoning engine that understands user inputs, leverages feedback loops, uses relevant contextual information from memory constructs and invokes plugins or tools to take goal-oriented action.
2. Tools, which are specialized components that perform a variety of tasks such as performing search, executing workflows, calling APIs, running code, calling LLMs for tasks such as summarization, and many more.
3. Reasoning and feedback loops to iterate
   1. Internal planning iteration loop to identify the most useful course of action
   2. Internal execution iteration loop to execute the plan step-by-step and assess what to do next at each stage
   3. User-facing feedback loop to communicate the thought process, seek confirmation and use user feedback to take the next step
4. Different types of “Memory” constructs to ensure that any decisions or predictions made by the reasoning engine are using accurate and contextually relevant information:
   1. Knowledge of the content, entities and terminology used by the organization - Needed to understand the semantics of the conversation with the user
   2. Awareness of the current context in the conversation with the user, i.e. what questions and answers have been exchanged and what decisions have been made already - Needed to have a true interactive conversation with the user where every response is unique and tailored to the context
   3. Knowledge of the tasks that can be performed in the environment for the user and what business processes or rules should be followed - Needed to select the right tools for the request and how to apply them
   4. Awareness of what operations and processes are in progress, and which stage of completion they are at - Needed to make sure that multi-step synchronous and asynchronous processes are tracked and driven to completion
5. Steerability tools to adapt out-of-the-box behavior based on customer needs
6. Governance mechanisms to ensure safety and accuracy of Assistant interactions and provide stakeholders visibility into what users are asking for and whether they are getting the assistance they need

Learn more these components and more

• Reasoning Loops in the AI Assistant
• Memory Constructs and Context Management
• Guardrails in the AI Assistant
• Steerability Tools
• How the Reasoning Engine Works
• LLM Fundamentals