Bhaskarjit Sarmah

It's the Friday afternoon task every analyst knows. You've been handed a messy folder full of Excel files, and by the end of the day, your manager expects a clean, board-ready presentation. 

In this live challenge, three panelists each work with a different AI coworker. Starting with the exact same folder and the same business brief, they have about 15 minutes to complete the same task using nothing but plain-English instructions. 

The task: Organize the folder, identify and clean the relevant data, analyze the sales workbook, reconcile the numbers against a target finance figure, and turn the results into a polished executive presentation. Along the way, the AI coworker must make sensible decisions, ask for confirmation before destructive actions like deleting files, handle ambiguous requests, and explain what it cannot determine instead of making things up. 

Every step happens live. You'll watch each AI coworker inspect and organize files, clean messy spreadsheets, work through incomplete and imperfect data, create charts, and build a leadership-ready presentation. Once all three have finished, the panelists compare their experiences and discuss where each AI coworker succeeded, where it struggled, and when human oversight was still essential. 

The question for the room is simple: Are today's AI coworkers ready to independently handle real business workflows, or do they still need a human guiding them every step of the way?

WHAT YOU'LL SEE 

• Three AI coworkers tackle the exact same business task, starting with the same messy folder and the same brief.  
• A real workflow unfold live, from organizing files and cleaning spreadsheets to building a board-ready presentation.  

• How each AI coworker interprets plain-English instructions, handles ambiguous requests, and makes decisions when the data isn't straightforward.  

• Trust-critical moments in action, including asking for confirmation before deleting files, reconciling numbers against finance targets, and flagging missing information instead of guessing.  

• A side-by-side comparison of each AI coworker's reasoning, reliability, and quality of output. 

In this hands-on workshop, participants will learn how reinforcement learning (RL) is used to train large language model–based agents that can make sequential decisions, interact with environments, call tools autonomously, and improve performance through experience. 
 
We will cover RL fundamentals for LLM agents, extend Markov Decision Processes (MDPs) to agent settings, explore modular RL frameworks, and dive into practical implementations using OpenPipe’s Agent Reinforcement Trainer (ART). By the end, attendees will understand how to design, train, and evaluate RL-based LLM agents for real-world tasks. 

*Note: These are tentative details and are subject to change.