Target Audience

• Product Managers
• Data & Analytics professionals
• Software Engineers
• Techn Consultants
• Technical Decision-Makers seeking to design AI-driven systems that reliably achieve complex goals using modern large language models

Description

Large Language Models are already highly capable. They can reason, decompose tasks, synthesize information, and often exceed our expectations. 

The real challenge is no longer what models can do, but how to consistently reach a required outcome despite known constraints. 

This workshop introduces Agentic Thinking: a capability-first methodology for designing AI systems that operate autonomously while remaining aligned with clear goals, constraints, and expectations. 

Main Topics

Agentic Systems by First Principles 

• From single-response models to goal-driven systems 

• Prompt vs workflow vs agent: functional distinctions 

• Agents as a way to organize strong capabilities over time 

• When an agent is the right abstraction and when it isn’t 

Outcome Design 

• Answers vs outcomes vs repeatable behavior 

• Defining success in a way models can converge toward 

• Behavioral outcomes rather than textual outputs 

• Avoiding over-specification and under-specification 

Task Decomposition & Guardrails 

• When models should decompose tasks autonomously 

• When task structure must be external 

• Single-agent multi-step vs multi-role agent designs 

Guardrails as performance enablers: 

• Guardrails in prompts 

• Guardrails at the tool and capability level 

• Role-based and agent-level guardrails 

• Process-level guardrails and decision boundaries 

Human-in-the-Process 

• Human-in-the-loop vs human-in-the-process 

• Why constant approval slows capable systems 

• Strategic human touchpoints in agentic workflows 

• Human judgment as acceleration, not correction 

Trajectory Control 

• Why “run to completion” often fails 

• Guiding agent behavior during execution 

• Checkpoints, decision moments, and course correction 

• Trajectory control vs guardrails 

• Maintaining alignment without micromanagement 

Stability Under Change 

• Why successful demos fail in real usage 

• Identifying critical assumptions 

• Outcome stability under changing context and scale 

• Evaluating agentic designs beyond first success 

Agentic Thinking as a Design Framework 

• The Agentic Design Map 

• Key design questions before granting autonomy 

• Applying agentic thinking across teams and products 

• Translating the methodology into real organizational decisions 

Key Learning Outcomes

At the end of the workshop, you will:

• Understand what AI agents are and how they differ from prompts and automated workflows 
• Learn how to define outcomes that AI systems can reliably converge toward 
• Explore task decomposition strategies that balance flexibility and control 
• Design guardrails at multiple levels to enable performance rather than restrict it 
• Understand the difference between human-in-the-loop and human-in-the-process participation 
• Learn how to guide agent behavior during execution, not only at design time 
• Evaluate whether an agentic solution remains stable under changing assumptions 
• Leave with a reusable mental model and concrete design artifacts for future projects