The next phase of AI agent design is cognitive
Most AI programs begin with a visible result: a ticket update, customer reply or approval note. These outputs are easy to identify and hand to a model. But they rarely represent the full work being performed.
Consider a senior support agent responding to a billing complaint. The visible output is a polished reply. Behind it sits something more valuable: recognising a repeated failure, recalling a previous commitment, sensing relationship risk and deciding when to depart from the standard process.
Automate the reply without capturing that thinking, and you may make the work faster while making the outcome worse.
In this two-hour lab, you’ll learn how to identify those hidden contributions and design them deliberately into your agent through context, tools, guardrails and human escalation.
You’ll leave with:
- A practical method for identifying hidden human judgment and context
- A structured canvas for turning those insights into agent requirements
- A blueprint covering prompts, grounding, tools, guardrails and escalation
- Evaluation cases for testing whether your agent preserves what matters
This lab is designed for teams building AI agents or deciding where AI should fit within their operations.
Speaker
