Skip to main content

Facilitator Guide

Agentic AI

What’s covered

  • How agentic AI differs from standard generative AI tools
  • How agentic systems plan, act, and iterate to complete multi-step tasks
  • Benefits of agentic AI, including efficiency, consistency, and scale
  • Key risks: verification challenges, agentic looping, bias amplification, and data security
  • Supervisory responsibilities and practical best practices for lawyers
  • Questions to ask when evaluating or deploying agentic tools

Before the session

Prepare the Attendees

Send the attendees an invitation for the session. Include this link, which has the videos and exercise they’ll need to prepare.

https://www.hotshotlegal.com/discussions/agentic-ai/attendee

Prepare Yourself

Watch the videos and read the exercise so you’re familiar with the Hotshot material.

  • still frame from Agentic AI

    Course

    Agentic AI

    An overview of agentic AI systems that can plan and execute multi-step legal workflows autonomously. Covers how agentic AI differs from basic GenAI chatbots, practical applications in legal practice, key risks including verification challenges and bias amplification, and best practices for delegating work to AI agents while maintaining professional responsibility.

  • Icon of a document

    Exercise

    PDF

During the session

Part 1: Knowledge Check

Ask attendees these questions to ensure they understand the topic:

  • How does agentic AI differ from a standard generative AI chatbot? What can it do that a standard tool cannot?
  • What does the plan-act-evaluate-adjust cycle mean in practice, and why does it matter for how lawyers supervise this work?
  • What is agentic looping, and what steps can you take to prevent or catch it?
  • How can bias get amplified in an agentic workflow in ways it wouldn’t in a single-prompt interaction?
  • What questions should you ask before deploying an agentic system on a matter?
  • How does a lawyer’s supervisory responsibility apply when an agentic system is handling multiple steps independently?

Part 2: Group Exercise

Lead a group discussion based on the exercise. To encourage collaboration, divide the attendees into groups to discuss the exercise, then have a representative from each group summarize their views to the larger group. Call on people to share their thoughts and ask others to respond. 

Tip for remote sessions: use your web conferencing system’s breakout room feature to divide people into groups.

FACILITATOR NOTES

The exercise is designed to surface the key tensions in deploying agentic AI responsibly—there are no strictly right answers. Use these notes to help guide the discussion.

Pros: The task is well-suited for agentic AI—repetitive, high-volume, and clearly defined. The client is asking about AI, which opens the door for a productive conversation about how the firm is using it.

Cons: Ongoing engagements create a risk of over-reliance. If the system runs each week without rigorous review, errors can accumulate before anyone catches them. The ambiguous firm policy also means there is no clear guidance on how to supervise a multi-step workflow or what the agent is permitted to access.

Push attendees to think concretely about supervision—not just “I’ll review the output,” but what specifically they would check, at what points, and how they would frame the task to make that review meaningful.

Part 3: Firm Specifics

Share your own experiences and guidance, including:

  • General practice tips (dos and don’ts)
  • Firm-specific guidance and practices, including your firm’s AI use policy and any existing guidance on supervising agentic workflows or communicating with clients about AI use

After the session

To continue their learning, attendees can check out related courses in Hotshot’s Artificial Intelligence topic.