Does It Actually Work?
Evaluating, Red-Teaming & Governing AI Agents Before They Reach Production
A high-impact practical workshop on testing and securing AI agents before they transition from impressive prototypes to autonomous real-world systems.
Ass. Prof. Oussama H. Hamid
Higher Colleges of Technology (HCT), Abu Dhabi, UAE
Workshop Syllabus & Technical Modules
Comprehensive hands-on curriculum structured to turn prototype AI agents into secure, production-grade systems.
Agent Architecture & Planning
- Deconstructing autonomous reasoning & execution loops
- Tool integration APIs and structured function calling
- Multi-agent coordination & context degradation risks
Empirical Agent Evaluation
- Beyond static benchmarks: measuring task execution success
- Testing behavior against ambiguous or incomplete prompts
- Evaluating resilience in non-deterministic environments
Red-Teaming & Adversarial Attack
- Indirect prompt injection in RAG & vector stores
- Malicious tool payloads and unauthorized API exploitation
- Simulating rogue agent behaviors and goal hijacking
Tool Access & Permission Control
- Principle of least privilege for autonomous agent tools
- Sandboxing code execution and API authorization filters
- Preventing unauthorized database reads/writes
Governance & Human-in-the-Loop
- Establishing mandatory human approval thresholds
- Compliance frameworks for enterprise AI deployment
- Comprehensive audit logging for agent actions
Pre-Production Readiness
- Go/No-Go pre-deployment evaluation checklist
- Real-time agent telemetry, fallback & monitoring
- Building continuous safety mechanisms post-launch
The Agentic Shift Requires New Testing Paradigms
AI agents are no longer limited to generating text. They can reason, plan, access tools, retrieve data, make decisions and execute multi-step tasks.
Key Takeaway
That changes how we test, secure and govern AI systems before they reach production.
Autonomous Agent Execution Loop
The Agent Readiness Journey
From prototype experimentation to monitored production deployment.
Understand
Evaluate
Red-Team
Secure
Govern
Deploy
01 Understand
Milestone: Prototype02 Evaluate
Milestone: Testing03 Red-Team
Milestone: Security04 Secure
Milestone: Human Approval05 Govern
Milestone: Monitoring06 Deploy
Milestone: ProductionBefore an AI Agent Goes Live...
Core operational challenges and stress tests addressed during evaluation:
Does it still work outside a controlled demo?
What happens with ambiguous instructions?
Can adversarial prompts manipulate behavior?
Can malicious tool inputs change execution?
What data/systems can the agent access?
When should a human approve an action?
How do we detect when the agent fails?
What metrics should be monitored post-launch?
Leave With a Practical AI-Agent Readiness Framework
Who Should Attend?
Designed for professionals and researchers building, evaluating or deploying agentic AI systems.